Facial Expression Recognition¶
1. Introduction¶
Facial emotion recognition is one of the most interesting research areas in which artificial intelligence and human psychology are combined to understand emotions through facial expressions. This project focuses on analyzing and classifying emotions like happiness, sadness, anger, neutral, disgust, contempt, fear, and surprise by interpreting facial features using advanced algorithms.¶
What's the main topic?¶
Our topic is Facial Emotion Recognition, which involves the identification and classification of emotions using facial expressions through machine learning and deep learning techniques. This technology aspires to bridge the gap between human emotions and artificial intelligence, making systems understand and respond to human feelings in a more empathetic and personalized manner. With its wide range of applications, including healthcare, customer service, and security, facial emotion recognition can enhance user experiences by providing more tailored and responsive interactions.¶
What questions are we trying to answer with our analysis?¶
The questions that this project will answer are as follows:¶
- How can facial emotion recognition systems be integrated into everyday AI applications to make them more responsive to human emotions and needs?
- What are the contributions of emotion recognition to the improvement of user experience and personalization of interactions in healthcare, customer service, and entertainment?
- How can ethical issues of facial emotion recognition be addressed in order to make sure that this technology is used responsibly and does not infringe on privacy or promote bias?
Why are these questions significant?¶
Developing empathetic and emotionally aware AI is vital for improving human-computer interactions. Addressing ethical concerns and ensuring fairness will enhance user experiences in areas like mental health support, personalized customer service, and secure applications, enabling responsible deployment across diverse sectors.¶
Data preprocessing and Data Cleaning¶
In [2]:
# Importing the required Libraries
import os
import shutil
import pandas as pd
import numpy as np
from glob import glob
import matplotlib.pyplot as plt
from sklearn.metrics import f1_score
from sklearn.model_selection import StratifiedKFold
import hashlib
from PIL import Image
import torch
import torch.nn as nn
import torch.optim as optim
from torch.nn import functional as F
from torch.optim import lr_scheduler
import torchvision.transforms as T
from torch.utils.data import Dataset, DataLoader,SubsetRandomSampler,ConcatDataset
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator,img_to_array, load_img
In [3]:
train_path = r'C:\Users\Harshitha GR\Desktop\saee\train'
val_path = r'C:\Users\Harshitha GR\Desktop\saee\val'
This code initializes two ImageDataGenerator instances for handling training and validation image datasets in a deep learning context.¶
Both generators load images from specified directories, resize them to 48x48 pixels, and classify them as grayscale and categorical for multi-class classification tasks, preparing the data for a convolutional neural network (CNN) training pipeline.¶
In [4]:
# Creating ImageDataGenerator for training and validation
train_datagen = ImageDataGenerator(
rescale=1./255,
rotation_range=20,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
fill_mode='nearest'
)
val_datagen = ImageDataGenerator(rescale=1./255)
# Loading the training images
train_generator = train_datagen.flow_from_directory(
train_path,
target_size=(48, 48),
batch_size=128,
color_mode='grayscale', # Use grayscale since the images are grayscale
class_mode='categorical'
)
# Loading the validation images
validation_generator = val_datagen.flow_from_directory(
val_path,
target_size=(48, 48),
batch_size=128,
color_mode='grayscale',
class_mode='categorical'
)
Found 31259 images belonging to 8 classes. Found 8077 images belonging to 8 classes.
In [5]:
# Checking the number of classes and samples in the training and validation sets
print("Training Classes and Counts:")
for class_name, count in train_generator.class_indices.items():
print(f"{class_name}: {len(os.listdir(os.path.join(train_path, class_name)))}")
print("\nValidation Classes and Counts:")
for class_name, count in validation_generator.class_indices.items():
print(f"{class_name}: {len(os.listdir(os.path.join(val_path, class_name)))}")
Training Classes and Counts: ANGER: 3820 CONTEMPT: 2018 DISGUST: 2155 FEAR: 3896 HAPPINESS: 7069 NEUTRAL: 4853 SADNESS: 4747 SURPRISE: 2701 Validation Classes and Counts: ANGER: 977 CONTEMPT: 521 DISGUST: 561 FEAR: 1013 HAPPINESS: 1789 NEUTRAL: 1234 SADNESS: 1214 SURPRISE: 768
The provided code defines two functions for identifying duplicate images and unwanted files in specified directories. The hash_image function generates a unique hash for each image by resizing it and converting it to bytes. The check_duplicates_and_unwanted_files function traverses a directory, checks each file's extension, and uses the hash function to detect duplicates, storing paths of duplicates and unwanted files separately.¶
In [6]:
def hash_image(file_path):
with Image.open(file_path) as img:
img = img.convert('RGB').resize((48, 48))
img_bytes = img.tobytes()
return hashlib.md5(img_bytes).hexdigest()
def check_duplicates_and_unwanted_files(directory):
seen_hashes = set()
duplicate_files = []
unwanted_files = []
for subdir, _, files in os.walk(directory):
for file in files:
file_path = os.path.join(subdir, file)
if not file.lower().endswith(('.png', '.jpg', '.jpeg', '.gif')):
unwanted_files.append(file_path)
continue
try:
img_hash = hash_image(file_path)
if img_hash in seen_hashes:
duplicate_files.append(file_path)
else:
seen_hashes.add(img_hash)
except Exception as e:
print(f"Error processing {file_path}: {e}")
return duplicate_files, unwanted_files
# Checking duplicates and unwanted files in the training and validation directories
train_duplicates, train_unwanted = check_duplicates_and_unwanted_files(train_path)
val_duplicates, val_unwanted = check_duplicates_and_unwanted_files(val_path)
print(f"Training Duplicates: {len(train_duplicates)}")
print(f"\nTraining Unwanted Files:{len(train_unwanted)}")
print(f"\nValidation Duplicates:{len(val_duplicates)}")
print(f"\nValidation Unwanted Files:{len(val_unwanted)}")
Training Duplicates: 0 Training Unwanted Files:0 Validation Duplicates:0 Validation Unwanted Files:1
In [7]:
def remove_duplicates(duplicate_files):
"""Remove duplicate files from the dataset."""
for file_path in duplicate_files:
try:
os.remove(file_path)
#print(f"Removed duplicate: {file_path}")
except Exception as e:
print(f"Error removing {file_path}: {e}")
# Remove duplicates from the training set
array = [train_duplicates, train_unwanted, val_duplicates, val_unwanted]
for i in array:
remove_duplicates(i)
print("Duplicate images have been removed.")
print(f"Training Duplicates: {len(train_duplicates)}")
print(f"\nTraining Unwanted Files:{len(train_unwanted)}")
print(f"\nValidation Duplicates:{len(val_duplicates)}")
print(f"\nValidation Unwanted Files:{len(val_unwanted)}")
Duplicate images have been removed. Training Duplicates: 0 Training Unwanted Files:0 Validation Duplicates:0 Validation Unwanted Files:1
In [8]:
train_generator1 = train_datagen.flow_from_directory(
train_path,
target_size=(48, 48),
batch_size=128,
color_mode='rgb',
class_mode='categorical'
)
# Loading the validation images
validation_generator1 = val_datagen.flow_from_directory(
val_path,
target_size=(48, 48),
batch_size=64,
color_mode='rgb',
class_mode='categorical'
)
print(f"Number of training images: {train_generator1.samples}")
print(f"Number of validation images: {validation_generator1.samples}")
print(f"Number of classes: {len(train_generator1.class_indices)}")
print("Class Labels:")
for label, index in train_generator1.class_indices.items():
print(f"{index}: {label}")
Found 31259 images belonging to 8 classes. Found 8077 images belonging to 8 classes. Number of training images: 31259 Number of validation images: 8077 Number of classes: 8 Class Labels: 0: ANGER 1: CONTEMPT 2: DISGUST 3: FEAR 4: HAPPINESS 5: NEUTRAL 6: SADNESS 7: SURPRISE
In [9]:
def plot_sample_images(generator, class_labels, num_samples=5):
"""Plot sample images from each class."""
for class_label in class_labels:
plt.figure(figsize=(10, 5))
plt.suptitle(f"Samples from Class: {class_label}", fontsize=16)
# Get image paths for the class
class_dir = os.path.join(train_path, class_label)
image_files = os.listdir(class_dir)
for i in range(num_samples):
img_path = os.path.join(class_dir, image_files[i])
img = plt.imread(img_path)
plt.subplot(1, num_samples, i + 1)
plt.imshow(img, cmap='gray')
plt.axis('off')
plt.show()
# Plotting five sample images from each class
plot_sample_images(train_generator1.class_indices.keys(), train_generator1.class_indices.keys())
In [10]:
import matplotlib.pyplot as plt
class_counts = [len(os.listdir(os.path.join(train_path, class_name))) for class_name in train_generator1.class_indices.keys()]
plt.figure(figsize=(10, 6))
plt.bar(train_generator1.class_indices.keys(), class_counts)
plt.xlabel('Class')
plt.ylabel('Number of Images')
plt.title('Class Distribution in Training Set')
plt.show()
In [11]:
total_images = sum(class_counts)
class_distribution = {label: count / total_images * 100 for label, count in zip(train_generator1.class_indices.keys(), class_counts)}
print(class_distribution)
{'ANGER': 12.220480501615533, 'CONTEMPT': 6.455740746664961, 'DISGUST': 6.894014523817141, 'FEAR': 12.463610480181709, 'HAPPINESS': 22.61428708531943, 'NEUTRAL': 15.52512876291628, 'SADNESS': 15.186026424389777, 'SURPRISE': 8.640711475095172}
In [12]:
import matplotlib.pyplot as plt
class_counts = [len(os.listdir(os.path.join(val_path, class_name))) for class_name in validation_generator1.class_indices.keys()]
plt.figure(figsize=(10, 6))
plt.bar(validation_generator1.class_indices.keys(), class_counts)
plt.xlabel('Class')
plt.ylabel('Number of Images')
plt.title('Class Distribution in Validation Set')
plt.show()
Basic data exploration and summary statistics¶
In [13]:
# Function to calculate mean and std for images in a generator
def calculate_mean_std(generator):
mean = 0
std = 0
total_samples = 0
for images, _ in generator:
batch_samples = images.shape[0]
mean += images.mean() * batch_samples
std += images.std() * batch_samples
total_samples += batch_samples
# Breaking after one full epoch
if total_samples >= generator.samples:
break
mean /= total_samples
std /= total_samples
return mean, std
mean, std = calculate_mean_std(train_generator1)
print(f"Mean pixel intensity: {mean:.4f}")
print(f"Standard deviation of pixel intensity: {std:.4f}")
Mean pixel intensity: 0.5069 Standard deviation of pixel intensity: 0.2556
Setting up a Hypothesis¶
Null Hypothesis (𝐻o): The dataset is balanced across all classes.¶
Alternate Hypothesis (𝐻𝑎): The dataset is not balanced across classes, indicating that there is a significant difference in the number of instances among the classes.¶
The Chi-Square test is used to determine whether there is a significant association between the different classes in the dataset. If the Chi-Square test yields a low p-value (typically less than 0.05), it suggests that the distribution of images across classes is not random and that some classes may be over- or under-represented.
In [16]:
from scipy.stats import chisquare
import os
# Define the path to your training data
train_path = r'C:\Users\Harshitha GR\Desktop\saee\train'
# Step 1: Calculate observed frequencies (number of images in each class)
class_counts = {}
for class_name in os.listdir(train_path):
class_path = os.path.join(train_path, class_name)
if os.path.isdir(class_path):
class_counts[class_name] = len(os.listdir(class_path))
# Observed frequencies
observed_frequencies = list(class_counts.values())
print("Observed frequencies (number of images in each class):", observed_frequencies)
# Step 2: Calculate expected frequencies (average count across classes)
expected_frequency = sum(observed_frequencies) / len(observed_frequencies)
expected_frequencies = [expected_frequency] * len(observed_frequencies)
print("Expected frequencies:", expected_frequencies)
# Step 3: Perform Chi-Square Test
chi2_statistic, p_value = chisquare(f_obs=observed_frequencies, f_exp=expected_frequencies)
# Output results
print("Chi-Square Statistic:", chi2_statistic)
print("p-value:", p_value)
# Interpret the results
alpha = 0.05
if p_value < alpha:
print("Reject the Null Hypothesis (𝐻o): The dataset is NOT balanced across all classes.")
else:
print("Fail to Reject the Null Hypothesis (𝐻o): The dataset is balanced across all classes.")
Observed frequencies (number of images in each class): [3820, 2018, 2155, 3896, 7069, 4853, 4747, 2701] Expected frequencies: [3907.375, 3907.375, 3907.375, 3907.375, 3907.375, 3907.375, 3907.375, 3907.375] Chi-Square Statistic: 5041.417799673694 p-value: 0.0 Reject the Null Hypothesis (𝐻o): The dataset is NOT balanced across all classes.
1) When certain classes are under-represented, models may struggle to learn their characteristics effectively, leading to biased predictions and poor generalization on unseen data.¶
2) By augmenting the under-represented classes—through transformations such as rotation, flipping, and zooming—the dataset becomes more balanced. This enhances the model's ability to learn from a diverse range of examples within those classes, ultimately improving accuracy and robustness¶
In [18]:
from tensorflow.keras.preprocessing.image import ImageDataGenerator, img_to_array, load_img
import os
import numpy as np
# Define augmentation transformations
augmenter = ImageDataGenerator(
rotation_range=30,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
brightness_range=[0.8, 1.2]
)
# Define paths and classes
train_path = r'C:\Users\Harshitha GR\Desktop\saee\train' # Your training path
under_represented_classes_train = ["ANGER", "FEAR", "SADNESS", "NEUTRAL","CONTEMPT","DISGUST","SURPRISE"] # Classes to augment
augmented_path = r'C:\Users\Harshitha GR\Desktop\saee\augmented'
# Create directory for augmented images if it doesn’t exist
os.makedirs(augmented_path, exist_ok=True)
# Set the target count as the maximum number of images in the "HAPPINESS" class
target_class = "HAPPINESS" # The class with the most images
target_count = len(os.listdir(os.path.join(train_path, target_class)))
# Loop through each under-represented class and perform augmentation
for class_name in under_represented_classes_train:
class_path = os.path.join(train_path, class_name)
images = os.listdir(class_path)
current_class_count = len(images)
num_images_to_generate = target_count - current_class_count # Calculate how many images to generate
# Define path for augmented images of the current class
class_augmented_path = os.path.join(augmented_path, class_name)
os.makedirs(class_augmented_path, exist_ok=True)
generated_images = 0 # Counter for generated images
# Augment until the desired number of images is reached
while generated_images < num_images_to_generate:
for img_name in images:
img_path = os.path.join(class_path, img_name)
img = load_img(img_path, color_mode="grayscale", target_size=(48, 48))
img_array = img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
# Generate and save augmented images
for batch in augmenter.flow(img_array, batch_size=1, save_to_dir=class_augmented_path,
save_prefix='aug', save_format='png'):
generated_images += 1
if generated_images >= num_images_to_generate:
break
if generated_images >= num_images_to_generate:
break
print("Data augmentation complete for under-represented classes.")
Data augmentation complete for under-represented classes.
In [20]:
combined_data_path = r'C:\Users\Harshitha GR\Desktop\saee\combined_train_data'
# Step 1: Copying original images into the combined data directory
if not os.path.exists(combined_data_path):
os.makedirs(combined_data_path)
for class_name in os.listdir(train_path):
original_class_path = os.path.join(train_path, class_name)
combined_class_path = os.path.join(combined_data_path, class_name)
os.makedirs(combined_class_path, exist_ok=True)
for file_name in os.listdir(original_class_path):
file_path = os.path.join(original_class_path, file_name)
if os.path.isfile(file_path):
shutil.copy(file_path, combined_class_path)
# Step 2: Copy augmented images into the combined data directory
for class_name in os.listdir(augmented_path):
augmented_class_dir = os.path.join(augmented_path, class_name)
combined_class_dir = os.path.join(combined_data_path, class_name)
for file_name in os.listdir(augmented_class_dir):
file_path = os.path.join(augmented_class_dir, file_name)
if os.path.isfile(file_path):
shutil.copy(file_path, combined_class_dir)
print("Original and augmented data combined successfully.")
Original and augmented data combined successfully.
In [21]:
from tensorflow.keras.preprocessing.image import ImageDataGenerator, img_to_array, load_img
import os
import numpy as np
# Define augmentation transformations
augmenter = ImageDataGenerator(
rotation_range=30,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
brightness_range=[0.8, 1.2]
)
# Define paths and classes
val_path = r'C:\Users\Harshitha GR\Desktop\saee\val' # Your validation path
under_represented_classes_val = ["ANGER", "FEAR", "SADNESS", "NEUTRAL", "CONTEMPT", "DISGUST", "SURPRISE"] # Classes to augment
augmented_val_path = r'C:\Users\Harshitha GR\Desktop\saee\augmented_val'
# Create directory for augmented validation images if it doesn’t exist
os.makedirs(augmented_val_path, exist_ok=True)
# Set the target count as the maximum number of images in the "HAPPINESS" class for validation
val_target_class = "HAPPINESS" # The class with the most images
val_target_count = len(os.listdir(os.path.join(val_path, val_target_class)))
# Loop through each under-represented class and perform augmentation
for val_class_name in under_represented_classes_val:
val_class_path = os.path.join(val_path, val_class_name)
val_images = os.listdir(val_class_path)
current_val_class_count = len(val_images)
val_num_images_to_generate = val_target_count - current_val_class_count # Calculate how many images to generate
# Define path for augmented validation images of the current class
val_class_augmented_path = os.path.join(augmented_val_path, val_class_name)
os.makedirs(val_class_augmented_path, exist_ok=True)
val_generated_images = 0 # Counter for generated images
# Augment until the desired number of images is reached
while val_generated_images < val_num_images_to_generate:
for val_img_name in val_images:
val_img_path = os.path.join(val_class_path, val_img_name)
val_img = load_img(val_img_path, color_mode="grayscale", target_size=(48, 48))
val_img_array = img_to_array(val_img)
val_img_array = np.expand_dims(val_img_array, axis=0)
# Generate and save augmented images
for batch in augmenter.flow(val_img_array, batch_size=1, save_to_dir=val_class_augmented_path,
save_prefix='aug', save_format='png'):
val_generated_images += 1
if val_generated_images >= val_num_images_to_generate:
break
if val_generated_images >= val_num_images_to_generate:
break
print("Data augmentation complete for under-represented validation classes.")
Data augmentation complete for under-represented validation classes.
In [22]:
import os
import shutil
# Paths for validation data
val_path = r'C:\Users\Harshitha GR\Desktop\saee\val'
augmented_val_path = r'C:\Users\Harshitha GR\Desktop\saee\augmented_val'
combined_val_data_path = r'C:\Users\Harshitha GR\Desktop\saee\combined_val_data'
# Step 1: Copying original validation images into the combined data directory
if not os.path.exists(combined_val_data_path):
os.makedirs(combined_val_data_path)
for val_class_name in os.listdir(val_path):
original_val_class_path = os.path.join(val_path, val_class_name)
combined_val_class_path = os.path.join(combined_val_data_path, val_class_name)
os.makedirs(combined_val_class_path, exist_ok=True)
for val_file_name in os.listdir(original_val_class_path):
val_file_path = os.path.join(original_val_class_path, val_file_name)
if os.path.isfile(val_file_path):
shutil.copy(val_file_path, combined_val_class_path)
# Step 2: Copy augmented validation images into the combined data directory
for val_class_name in os.listdir(augmented_val_path):
augmented_val_class_dir = os.path.join(augmented_val_path, val_class_name)
combined_val_class_dir = os.path.join(combined_val_data_path, val_class_name)
for val_file_name in os.listdir(augmented_val_class_dir):
val_file_path = os.path.join(augmented_val_class_dir, val_file_name)
if os.path.isfile(val_file_path):
shutil.copy(val_file_path, combined_val_class_dir)
print("Original and augmented validation data combined successfully for validation set.")
Original and augmented validation data combined successfully for validation set.
In [23]:
train_datagen1 = ImageDataGenerator(
rescale=1.0/255,
rotation_range=30,
width_shift_range=0.3,
height_shift_range=0.3,
shear_range=0.3,
zoom_range=0.3,
horizontal_flip=True
)
train_generator2 = train_datagen1.flow_from_directory(
combined_data_path,
target_size=(48, 48),
color_mode="rgb",
batch_size=128,
class_mode="categorical",
shuffle=True
)
print("Training generator is ready with rgb images")
Found 52173 images belonging to 8 classes. Training generator is ready with rgb images
In [24]:
# Loading the validation images
validation_generator1 = val_datagen.flow_from_directory(
combined_val_data_path,
target_size=(48, 48),
batch_size=128,
color_mode='rgb',
class_mode='categorical'
)
print("Validation generator is ready with rgb images")
Found 14025 images belonging to 8 classes. Validation generator is ready with rgb images
In [25]:
# Checking the number of classes and samples in the training and validation sets
print("Training Classes and Counts:")
for class_name, count in train_generator2.class_indices.items():
print(f"{class_name}: {len(os.listdir(os.path.join(combined_data_path, class_name)))}")
Training Classes and Counts: ANGER: 6585 CONTEMPT: 5995 DISGUST: 6005 FEAR: 6632 HAPPINESS: 7069 NEUTRAL: 6822 SADNESS: 6830 SURPRISE: 6235
In [26]:
# Checking the number of classes and samples in the training and validation sets
print("Validation Classes and Counts:")
for class_name, count in validation_generator1.class_indices.items():
print(f"{class_name}: {len(os.listdir(os.path.join(val_path, class_name)))}")
Validation Classes and Counts: ANGER: 977 CONTEMPT: 521 DISGUST: 561 FEAR: 1013 HAPPINESS: 1789 NEUTRAL: 1234 SADNESS: 1214 SURPRISE: 768
In [27]:
import matplotlib.pyplot as plt
class_counts_new = [len(os.listdir(os.path.join(combined_data_path, class_name))) for class_name in train_generator2.class_indices.keys()]
plt.figure(figsize=(10, 6))
plt.bar(train_generator2.class_indices.keys(), class_counts_new)
plt.xlabel('Class')
plt.ylabel('Number of Images')
plt.title('Class Distribution in Training Set')
plt.show()
In [28]:
import matplotlib.pyplot as plt
class_counts_val = [len(os.listdir(os.path.join(combined_val_data_path, class_name))) for class_name in validation_generator1.class_indices.keys()]
plt.figure(figsize=(10, 6))
plt.bar(validation_generator1.class_indices.keys(), class_counts_val)
plt.xlabel('Class')
plt.ylabel('Number of Images')
plt.title('Class Distribution in Validation Set')
plt.show()
After comparing both graphs, we can see that the imbalance between the classes has reduced.¶
In [39]:
from tensorflow.keras.applications import VGG16
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.layers import Dense, Dropout, GlobalAveragePooling2D, Flatten
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
conv_base = VGG16(
weights='imagenet',
include_top = False,
input_shape=(48,48,3)
)
model = Sequential()
model.add(conv_base)
model.add(Flatten())
model.add(Dense(512,activation = 'relu'))
model.add(Dense(256,activation = 'relu'))
model.add(Dropout(0.4))
model.add(Dense(8,activation='softmax'))
# Compile the model
model.compile(
optimizer=Adam(learning_rate=0.00005),
loss='categorical_crossentropy',
metrics=['accuracy']
)
conv_base.summary()
Model: "vgg16"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩ │ input_layer_1 (InputLayer) │ (None, 48, 48, 3) │ 0 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block1_conv1 (Conv2D) │ (None, 48, 48, 64) │ 1,792 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block1_conv2 (Conv2D) │ (None, 48, 48, 64) │ 36,928 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block1_pool (MaxPooling2D) │ (None, 24, 24, 64) │ 0 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block2_conv1 (Conv2D) │ (None, 24, 24, 128) │ 73,856 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block2_conv2 (Conv2D) │ (None, 24, 24, 128) │ 147,584 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block2_pool (MaxPooling2D) │ (None, 12, 12, 128) │ 0 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block3_conv1 (Conv2D) │ (None, 12, 12, 256) │ 295,168 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block3_conv2 (Conv2D) │ (None, 12, 12, 256) │ 590,080 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block3_conv3 (Conv2D) │ (None, 12, 12, 256) │ 590,080 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block3_pool (MaxPooling2D) │ (None, 6, 6, 256) │ 0 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block4_conv1 (Conv2D) │ (None, 6, 6, 512) │ 1,180,160 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block4_conv2 (Conv2D) │ (None, 6, 6, 512) │ 2,359,808 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block4_conv3 (Conv2D) │ (None, 6, 6, 512) │ 2,359,808 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block4_pool (MaxPooling2D) │ (None, 3, 3, 512) │ 0 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block5_conv1 (Conv2D) │ (None, 3, 3, 512) │ 2,359,808 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block5_conv2 (Conv2D) │ (None, 3, 3, 512) │ 2,359,808 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block5_conv3 (Conv2D) │ (None, 3, 3, 512) │ 2,359,808 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ block5_pool (MaxPooling2D) │ (None, 1, 1, 512) │ 0 │ └─────────────────────────────────┴────────────────────────┴───────────────┘
Total params: 14,714,688 (56.13 MB)
Trainable params: 14,714,688 (56.13 MB)
Non-trainable params: 0 (0.00 B)
In [59]:
#Compile the model
model.compile(
optimizer=Adam(learning_rate=0.00005),
loss='categorical_crossentropy',
metrics=['accuracy']
)
# Define callbacks
callbacks = [
EarlyStopping(
monitor='val_loss',
patience=5,
restore_best_weights=True
),
ReduceLROnPlateau(
monitor='val_loss',
factor=0.2,
patience=3,
verbose=1
)
]
# Training the Model
history = model.fit(
train_generator2,
epochs=50,
validation_data=validation_generator1,
callbacks=callbacks
)
# Evaluate the model
loss, accuracy = model.evaluate(validation_generator1)
print(f"Validation Accuracy: {accuracy * 100:.2f}%")
# Save the trained model
model.save('VGG_50_EPOCHS.keras')
print("Model saved as 'VGG_50_EPOCHS.keras'.")
Epoch 1/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 513s 1s/step - accuracy: 0.7007 - loss: 0.7960 - val_accuracy: 0.5795 - val_loss: 1.1192 - learning_rate: 5.0000e-05 Epoch 2/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 514s 1s/step - accuracy: 0.7173 - loss: 0.7502 - val_accuracy: 0.5958 - val_loss: 1.0631 - learning_rate: 5.0000e-05 Epoch 3/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 515s 1s/step - accuracy: 0.7316 - loss: 0.7228 - val_accuracy: 0.5852 - val_loss: 1.0879 - learning_rate: 5.0000e-05 Epoch 4/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7360 - loss: 0.7044 - val_accuracy: 0.5996 - val_loss: 1.0586 - learning_rate: 5.0000e-05 Epoch 5/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 513s 1s/step - accuracy: 0.7442 - loss: 0.6868 - val_accuracy: 0.6098 - val_loss: 1.0387 - learning_rate: 5.0000e-05 Epoch 6/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 513s 1s/step - accuracy: 0.7462 - loss: 0.6757 - val_accuracy: 0.6297 - val_loss: 0.9823 - learning_rate: 5.0000e-05 Epoch 7/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 514s 1s/step - accuracy: 0.7559 - loss: 0.6577 - val_accuracy: 0.6229 - val_loss: 1.0104 - learning_rate: 5.0000e-05 Epoch 8/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 515s 1s/step - accuracy: 0.7591 - loss: 0.6493 - val_accuracy: 0.6257 - val_loss: 1.0030 - learning_rate: 5.0000e-05 Epoch 9/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 515s 1s/step - accuracy: 0.7653 - loss: 0.6374 - val_accuracy: 0.6419 - val_loss: 0.9445 - learning_rate: 5.0000e-05 Epoch 10/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7682 - loss: 0.6228 - val_accuracy: 0.6470 - val_loss: 0.9524 - learning_rate: 5.0000e-05 Epoch 11/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7719 - loss: 0.6127 - val_accuracy: 0.6592 - val_loss: 0.9255 - learning_rate: 5.0000e-05 Epoch 12/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7771 - loss: 0.5993 - val_accuracy: 0.6545 - val_loss: 0.9260 - learning_rate: 5.0000e-05 Epoch 13/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7819 - loss: 0.5911 - val_accuracy: 0.6604 - val_loss: 0.9341 - learning_rate: 5.0000e-05 Epoch 14/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7832 - loss: 0.5857 - val_accuracy: 0.6634 - val_loss: 0.9148 - learning_rate: 5.0000e-05 Epoch 15/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 512s 1s/step - accuracy: 0.7883 - loss: 0.5745 - val_accuracy: 0.6510 - val_loss: 0.9256 - learning_rate: 5.0000e-05 Epoch 16/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 513s 1s/step - accuracy: 0.7891 - loss: 0.5640 - val_accuracy: 0.6600 - val_loss: 0.9242 - learning_rate: 5.0000e-05 Epoch 17/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 501s 1s/step - accuracy: 0.7937 - loss: 0.5563 - val_accuracy: 0.6712 - val_loss: 0.8859 - learning_rate: 5.0000e-05 Epoch 18/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 502s 1s/step - accuracy: 0.7955 - loss: 0.5558 - val_accuracy: 0.6798 - val_loss: 0.8688 - learning_rate: 5.0000e-05 Epoch 19/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 501s 1s/step - accuracy: 0.7973 - loss: 0.5398 - val_accuracy: 0.6681 - val_loss: 0.9240 - learning_rate: 5.0000e-05 Epoch 20/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 496s 1s/step - accuracy: 0.8048 - loss: 0.5300 - val_accuracy: 0.6679 - val_loss: 0.9211 - learning_rate: 5.0000e-05 Epoch 21/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8056 - loss: 0.5249 - val_accuracy: 0.6849 - val_loss: 0.8673 - learning_rate: 5.0000e-05 Epoch 22/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 495s 1s/step - accuracy: 0.8094 - loss: 0.5192 - val_accuracy: 0.6738 - val_loss: 0.8970 - learning_rate: 5.0000e-05 Epoch 23/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8120 - loss: 0.5118 - val_accuracy: 0.6865 - val_loss: 0.8613 - learning_rate: 5.0000e-05 Epoch 24/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8147 - loss: 0.5047 - val_accuracy: 0.6771 - val_loss: 0.8773 - learning_rate: 5.0000e-05 Epoch 25/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8167 - loss: 0.4983 - val_accuracy: 0.6917 - val_loss: 0.8544 - learning_rate: 5.0000e-05 Epoch 26/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8186 - loss: 0.4912 - val_accuracy: 0.6860 - val_loss: 0.8806 - learning_rate: 5.0000e-05 Epoch 27/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 496s 1s/step - accuracy: 0.8225 - loss: 0.4803 - val_accuracy: 0.6902 - val_loss: 0.8736 - learning_rate: 5.0000e-05 Epoch 28/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 0s 1s/step - accuracy: 0.8246 - loss: 0.4742 Epoch 28: ReduceLROnPlateau reducing learning rate to 9.999999747378752e-06. 408/408 ━━━━━━━━━━━━━━━━━━━━ 497s 1s/step - accuracy: 0.8246 - loss: 0.4743 - val_accuracy: 0.6941 - val_loss: 0.8607 - learning_rate: 5.0000e-05 Epoch 29/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8376 - loss: 0.4409 - val_accuracy: 0.6996 - val_loss: 0.8611 - learning_rate: 1.0000e-05 Epoch 30/50 408/408 ━━━━━━━━━━━━━━━━━━━━ 494s 1s/step - accuracy: 0.8470 - loss: 0.4172 - val_accuracy: 0.6956 - val_loss: 0.8681 - learning_rate: 1.0000e-05 64/64 ━━━━━━━━━━━━━━━━━━━━ 17s 272ms/step - accuracy: 0.6903 - loss: 0.8589 Validation Accuracy: 69.17% Model saved as 'VGG_50_EPOCHS.keras'.
In [60]:
import matplotlib.pyplot as plt
# Retrieve metrics from the history object
train_loss = history.history['loss']
val_loss = history.history['val_loss']
train_accuracy = history.history['accuracy']
val_accuracy = history.history['val_accuracy']
epochs = range(1, len(train_loss) + 1)
# Plot Loss
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(epochs, train_loss, label='Training Loss', color='blue')
plt.plot(epochs, val_loss, label='Validation Loss', color='orange')
plt.title('Training and Validation Loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.grid(True)
# Plot Accuracy
plt.subplot(1, 2, 2)
plt.plot(epochs, train_accuracy, label='Training Accuracy', color='blue')
plt.plot(epochs, val_accuracy, label='Validation Accuracy', color='orange')
plt.title('Training and Validation Accuracy')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.grid(True)
# Show the plots
plt.tight_layout()
plt.show()
Transfer learning with ResNet50 as the base, adding custom fully connected layers, and compiling the model for 8-class image classification using categorical crossentropy and the Adam optimizer.¶
In [49]:
from tensorflow.keras.applications import ResNet50
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.layers import Dense, Dropout, GlobalAveragePooling2D, Flatten
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
conv_base = ResNet50(
weights='imagenet',
include_top=False,
input_shape=(48, 48, 3)
)
model_res = Sequential()
model_res.add(conv_base)
model_res.add(Flatten())
model_res.add(Dense(256,activation = 'relu'))
model_res.add(Dense(512,activation = 'relu'))
model_res.add(Dropout(0.45))
model_res.add(Dense(8,activation='softmax'))
# Compile the model
model_res.compile(
optimizer=Adam(learning_rate=0.0001),
loss='categorical_crossentropy',
metrics=['accuracy']
)
conv_base.summary()
Model: "resnet50"
┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ Connected to ┃ ┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩ │ input_layer_3 │ (None, 48, 48, 3) │ 0 │ - │ │ (InputLayer) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv1_pad │ (None, 54, 54, 3) │ 0 │ input_layer_3[0]… │ │ (ZeroPadding2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv1_conv (Conv2D) │ (None, 24, 24, │ 9,472 │ conv1_pad[0][0] │ │ │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv1_bn │ (None, 24, 24, │ 256 │ conv1_conv[0][0] │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv1_relu │ (None, 24, 24, │ 0 │ conv1_bn[0][0] │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ pool1_pad │ (None, 26, 26, │ 0 │ conv1_relu[0][0] │ │ (ZeroPadding2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ pool1_pool │ (None, 12, 12, │ 0 │ pool1_pad[0][0] │ │ (MaxPooling2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_1_conv │ (None, 12, 12, │ 4,160 │ pool1_pool[0][0] │ │ (Conv2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_1_bn │ (None, 12, 12, │ 256 │ conv2_block1_1_c… │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_1_relu │ (None, 12, 12, │ 0 │ conv2_block1_1_b… │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_2_conv │ (None, 12, 12, │ 36,928 │ conv2_block1_1_r… │ │ (Conv2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_2_bn │ (None, 12, 12, │ 256 │ conv2_block1_2_c… │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_2_relu │ (None, 12, 12, │ 0 │ conv2_block1_2_b… │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_0_conv │ (None, 12, 12, │ 16,640 │ pool1_pool[0][0] │ │ (Conv2D) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_3_conv │ (None, 12, 12, │ 16,640 │ conv2_block1_2_r… │ │ (Conv2D) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_0_bn │ (None, 12, 12, │ 1,024 │ conv2_block1_0_c… │ │ (BatchNormalizatio… │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_3_bn │ (None, 12, 12, │ 1,024 │ conv2_block1_3_c… │ │ (BatchNormalizatio… │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_add │ (None, 12, 12, │ 0 │ conv2_block1_0_b… │ │ (Add) │ 256) │ │ conv2_block1_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block1_out │ (None, 12, 12, │ 0 │ conv2_block1_add… │ │ (Activation) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_1_conv │ (None, 12, 12, │ 16,448 │ conv2_block1_out… │ │ (Conv2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_1_bn │ (None, 12, 12, │ 256 │ conv2_block2_1_c… │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_1_relu │ (None, 12, 12, │ 0 │ conv2_block2_1_b… │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_2_conv │ (None, 12, 12, │ 36,928 │ conv2_block2_1_r… │ │ (Conv2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_2_bn │ (None, 12, 12, │ 256 │ conv2_block2_2_c… │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_2_relu │ (None, 12, 12, │ 0 │ conv2_block2_2_b… │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_3_conv │ (None, 12, 12, │ 16,640 │ conv2_block2_2_r… │ │ (Conv2D) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_3_bn │ (None, 12, 12, │ 1,024 │ conv2_block2_3_c… │ │ (BatchNormalizatio… │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_add │ (None, 12, 12, │ 0 │ conv2_block1_out… │ │ (Add) │ 256) │ │ conv2_block2_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block2_out │ (None, 12, 12, │ 0 │ conv2_block2_add… │ │ (Activation) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_1_conv │ (None, 12, 12, │ 16,448 │ conv2_block2_out… │ │ (Conv2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_1_bn │ (None, 12, 12, │ 256 │ conv2_block3_1_c… │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_1_relu │ (None, 12, 12, │ 0 │ conv2_block3_1_b… │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_2_conv │ (None, 12, 12, │ 36,928 │ conv2_block3_1_r… │ │ (Conv2D) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_2_bn │ (None, 12, 12, │ 256 │ conv2_block3_2_c… │ │ (BatchNormalizatio… │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_2_relu │ (None, 12, 12, │ 0 │ conv2_block3_2_b… │ │ (Activation) │ 64) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_3_conv │ (None, 12, 12, │ 16,640 │ conv2_block3_2_r… │ │ (Conv2D) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_3_bn │ (None, 12, 12, │ 1,024 │ conv2_block3_3_c… │ │ (BatchNormalizatio… │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_add │ (None, 12, 12, │ 0 │ conv2_block2_out… │ │ (Add) │ 256) │ │ conv2_block3_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv2_block3_out │ (None, 12, 12, │ 0 │ conv2_block3_add… │ │ (Activation) │ 256) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_1_conv │ (None, 6, 6, 128) │ 32,896 │ conv2_block3_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_1_bn │ (None, 6, 6, 128) │ 512 │ conv3_block1_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_1_relu │ (None, 6, 6, 128) │ 0 │ conv3_block1_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_2_conv │ (None, 6, 6, 128) │ 147,584 │ conv3_block1_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_2_bn │ (None, 6, 6, 128) │ 512 │ conv3_block1_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_2_relu │ (None, 6, 6, 128) │ 0 │ conv3_block1_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_0_conv │ (None, 6, 6, 512) │ 131,584 │ conv2_block3_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_3_conv │ (None, 6, 6, 512) │ 66,048 │ conv3_block1_2_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_0_bn │ (None, 6, 6, 512) │ 2,048 │ conv3_block1_0_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_3_bn │ (None, 6, 6, 512) │ 2,048 │ conv3_block1_3_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_add │ (None, 6, 6, 512) │ 0 │ conv3_block1_0_b… │ │ (Add) │ │ │ conv3_block1_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block1_out │ (None, 6, 6, 512) │ 0 │ conv3_block1_add… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_1_conv │ (None, 6, 6, 128) │ 65,664 │ conv3_block1_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_1_bn │ (None, 6, 6, 128) │ 512 │ conv3_block2_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_1_relu │ (None, 6, 6, 128) │ 0 │ conv3_block2_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_2_conv │ (None, 6, 6, 128) │ 147,584 │ conv3_block2_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_2_bn │ (None, 6, 6, 128) │ 512 │ conv3_block2_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_2_relu │ (None, 6, 6, 128) │ 0 │ conv3_block2_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_3_conv │ (None, 6, 6, 512) │ 66,048 │ conv3_block2_2_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_3_bn │ (None, 6, 6, 512) │ 2,048 │ conv3_block2_3_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_add │ (None, 6, 6, 512) │ 0 │ conv3_block1_out… │ │ (Add) │ │ │ conv3_block2_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block2_out │ (None, 6, 6, 512) │ 0 │ conv3_block2_add… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_1_conv │ (None, 6, 6, 128) │ 65,664 │ conv3_block2_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_1_bn │ (None, 6, 6, 128) │ 512 │ conv3_block3_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_1_relu │ (None, 6, 6, 128) │ 0 │ conv3_block3_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_2_conv │ (None, 6, 6, 128) │ 147,584 │ conv3_block3_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_2_bn │ (None, 6, 6, 128) │ 512 │ conv3_block3_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_2_relu │ (None, 6, 6, 128) │ 0 │ conv3_block3_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_3_conv │ (None, 6, 6, 512) │ 66,048 │ conv3_block3_2_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_3_bn │ (None, 6, 6, 512) │ 2,048 │ conv3_block3_3_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_add │ (None, 6, 6, 512) │ 0 │ conv3_block2_out… │ │ (Add) │ │ │ conv3_block3_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block3_out │ (None, 6, 6, 512) │ 0 │ conv3_block3_add… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_1_conv │ (None, 6, 6, 128) │ 65,664 │ conv3_block3_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_1_bn │ (None, 6, 6, 128) │ 512 │ conv3_block4_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_1_relu │ (None, 6, 6, 128) │ 0 │ conv3_block4_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_2_conv │ (None, 6, 6, 128) │ 147,584 │ conv3_block4_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_2_bn │ (None, 6, 6, 128) │ 512 │ conv3_block4_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_2_relu │ (None, 6, 6, 128) │ 0 │ conv3_block4_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_3_conv │ (None, 6, 6, 512) │ 66,048 │ conv3_block4_2_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_3_bn │ (None, 6, 6, 512) │ 2,048 │ conv3_block4_3_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_add │ (None, 6, 6, 512) │ 0 │ conv3_block3_out… │ │ (Add) │ │ │ conv3_block4_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv3_block4_out │ (None, 6, 6, 512) │ 0 │ conv3_block4_add… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_1_conv │ (None, 3, 3, 256) │ 131,328 │ conv3_block4_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_1_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block1_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_1_relu │ (None, 3, 3, 256) │ 0 │ conv4_block1_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_2_conv │ (None, 3, 3, 256) │ 590,080 │ conv4_block1_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_2_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block1_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_2_relu │ (None, 3, 3, 256) │ 0 │ conv4_block1_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_0_conv │ (None, 3, 3, │ 525,312 │ conv3_block4_out… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_3_conv │ (None, 3, 3, │ 263,168 │ conv4_block1_2_r… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_0_bn │ (None, 3, 3, │ 4,096 │ conv4_block1_0_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_3_bn │ (None, 3, 3, │ 4,096 │ conv4_block1_3_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_add │ (None, 3, 3, │ 0 │ conv4_block1_0_b… │ │ (Add) │ 1024) │ │ conv4_block1_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block1_out │ (None, 3, 3, │ 0 │ conv4_block1_add… │ │ (Activation) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_1_conv │ (None, 3, 3, 256) │ 262,400 │ conv4_block1_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_1_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block2_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_1_relu │ (None, 3, 3, 256) │ 0 │ conv4_block2_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_2_conv │ (None, 3, 3, 256) │ 590,080 │ conv4_block2_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_2_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block2_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_2_relu │ (None, 3, 3, 256) │ 0 │ conv4_block2_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_3_conv │ (None, 3, 3, │ 263,168 │ conv4_block2_2_r… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_3_bn │ (None, 3, 3, │ 4,096 │ conv4_block2_3_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_add │ (None, 3, 3, │ 0 │ conv4_block1_out… │ │ (Add) │ 1024) │ │ conv4_block2_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block2_out │ (None, 3, 3, │ 0 │ conv4_block2_add… │ │ (Activation) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_1_conv │ (None, 3, 3, 256) │ 262,400 │ conv4_block2_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_1_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block3_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_1_relu │ (None, 3, 3, 256) │ 0 │ conv4_block3_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_2_conv │ (None, 3, 3, 256) │ 590,080 │ conv4_block3_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_2_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block3_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_2_relu │ (None, 3, 3, 256) │ 0 │ conv4_block3_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_3_conv │ (None, 3, 3, │ 263,168 │ conv4_block3_2_r… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_3_bn │ (None, 3, 3, │ 4,096 │ conv4_block3_3_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_add │ (None, 3, 3, │ 0 │ conv4_block2_out… │ │ (Add) │ 1024) │ │ conv4_block3_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block3_out │ (None, 3, 3, │ 0 │ conv4_block3_add… │ │ (Activation) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_1_conv │ (None, 3, 3, 256) │ 262,400 │ conv4_block3_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_1_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block4_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_1_relu │ (None, 3, 3, 256) │ 0 │ conv4_block4_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_2_conv │ (None, 3, 3, 256) │ 590,080 │ conv4_block4_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_2_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block4_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_2_relu │ (None, 3, 3, 256) │ 0 │ conv4_block4_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_3_conv │ (None, 3, 3, │ 263,168 │ conv4_block4_2_r… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_3_bn │ (None, 3, 3, │ 4,096 │ conv4_block4_3_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_add │ (None, 3, 3, │ 0 │ conv4_block3_out… │ │ (Add) │ 1024) │ │ conv4_block4_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block4_out │ (None, 3, 3, │ 0 │ conv4_block4_add… │ │ (Activation) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_1_conv │ (None, 3, 3, 256) │ 262,400 │ conv4_block4_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_1_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block5_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_1_relu │ (None, 3, 3, 256) │ 0 │ conv4_block5_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_2_conv │ (None, 3, 3, 256) │ 590,080 │ conv4_block5_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_2_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block5_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_2_relu │ (None, 3, 3, 256) │ 0 │ conv4_block5_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_3_conv │ (None, 3, 3, │ 263,168 │ conv4_block5_2_r… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_3_bn │ (None, 3, 3, │ 4,096 │ conv4_block5_3_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_add │ (None, 3, 3, │ 0 │ conv4_block4_out… │ │ (Add) │ 1024) │ │ conv4_block5_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block5_out │ (None, 3, 3, │ 0 │ conv4_block5_add… │ │ (Activation) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_1_conv │ (None, 3, 3, 256) │ 262,400 │ conv4_block5_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_1_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block6_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_1_relu │ (None, 3, 3, 256) │ 0 │ conv4_block6_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_2_conv │ (None, 3, 3, 256) │ 590,080 │ conv4_block6_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_2_bn │ (None, 3, 3, 256) │ 1,024 │ conv4_block6_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_2_relu │ (None, 3, 3, 256) │ 0 │ conv4_block6_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_3_conv │ (None, 3, 3, │ 263,168 │ conv4_block6_2_r… │ │ (Conv2D) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_3_bn │ (None, 3, 3, │ 4,096 │ conv4_block6_3_c… │ │ (BatchNormalizatio… │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_add │ (None, 3, 3, │ 0 │ conv4_block5_out… │ │ (Add) │ 1024) │ │ conv4_block6_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv4_block6_out │ (None, 3, 3, │ 0 │ conv4_block6_add… │ │ (Activation) │ 1024) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_1_conv │ (None, 2, 2, 512) │ 524,800 │ conv4_block6_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_1_bn │ (None, 2, 2, 512) │ 2,048 │ conv5_block1_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_1_relu │ (None, 2, 2, 512) │ 0 │ conv5_block1_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_2_conv │ (None, 2, 2, 512) │ 2,359,808 │ conv5_block1_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_2_bn │ (None, 2, 2, 512) │ 2,048 │ conv5_block1_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_2_relu │ (None, 2, 2, 512) │ 0 │ conv5_block1_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_0_conv │ (None, 2, 2, │ 2,099,200 │ conv4_block6_out… │ │ (Conv2D) │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_3_conv │ (None, 2, 2, │ 1,050,624 │ conv5_block1_2_r… │ │ (Conv2D) │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_0_bn │ (None, 2, 2, │ 8,192 │ conv5_block1_0_c… │ │ (BatchNormalizatio… │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_3_bn │ (None, 2, 2, │ 8,192 │ conv5_block1_3_c… │ │ (BatchNormalizatio… │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_add │ (None, 2, 2, │ 0 │ conv5_block1_0_b… │ │ (Add) │ 2048) │ │ conv5_block1_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block1_out │ (None, 2, 2, │ 0 │ conv5_block1_add… │ │ (Activation) │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_1_conv │ (None, 2, 2, 512) │ 1,049,088 │ conv5_block1_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_1_bn │ (None, 2, 2, 512) │ 2,048 │ conv5_block2_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_1_relu │ (None, 2, 2, 512) │ 0 │ conv5_block2_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_2_conv │ (None, 2, 2, 512) │ 2,359,808 │ conv5_block2_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_2_bn │ (None, 2, 2, 512) │ 2,048 │ conv5_block2_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_2_relu │ (None, 2, 2, 512) │ 0 │ conv5_block2_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_3_conv │ (None, 2, 2, │ 1,050,624 │ conv5_block2_2_r… │ │ (Conv2D) │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_3_bn │ (None, 2, 2, │ 8,192 │ conv5_block2_3_c… │ │ (BatchNormalizatio… │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_add │ (None, 2, 2, │ 0 │ conv5_block1_out… │ │ (Add) │ 2048) │ │ conv5_block2_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block2_out │ (None, 2, 2, │ 0 │ conv5_block2_add… │ │ (Activation) │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_1_conv │ (None, 2, 2, 512) │ 1,049,088 │ conv5_block2_out… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_1_bn │ (None, 2, 2, 512) │ 2,048 │ conv5_block3_1_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_1_relu │ (None, 2, 2, 512) │ 0 │ conv5_block3_1_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_2_conv │ (None, 2, 2, 512) │ 2,359,808 │ conv5_block3_1_r… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_2_bn │ (None, 2, 2, 512) │ 2,048 │ conv5_block3_2_c… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_2_relu │ (None, 2, 2, 512) │ 0 │ conv5_block3_2_b… │ │ (Activation) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_3_conv │ (None, 2, 2, │ 1,050,624 │ conv5_block3_2_r… │ │ (Conv2D) │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_3_bn │ (None, 2, 2, │ 8,192 │ conv5_block3_3_c… │ │ (BatchNormalizatio… │ 2048) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_add │ (None, 2, 2, │ 0 │ conv5_block2_out… │ │ (Add) │ 2048) │ │ conv5_block3_3_b… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ conv5_block3_out │ (None, 2, 2, │ 0 │ conv5_block3_add… │ │ (Activation) │ 2048) │ │ │ └─────────────────────┴───────────────────┴────────────┴───────────────────┘
Total params: 23,587,712 (89.98 MB)
Trainable params: 23,534,592 (89.78 MB)
Non-trainable params: 53,120 (207.50 KB)
In [55]:
#Compile the model
model_res.compile(
optimizer=Adam(learning_rate=0.00005), # Small learning rate for stability
loss='categorical_crossentropy', # Suitable for multi-class classification
metrics=['accuracy']
)
# Define callbacks
callbacks = [
EarlyStopping(
monitor='val_loss',
patience=5,
restore_best_weights=True
),
ReduceLROnPlateau(
monitor='val_loss',
factor=0.2,
patience=3,
verbose=1
)
]
# Training the Model
history = model_res.fit(
train_generator2,
epochs=30,
validation_data=validation_generator1,
callbacks=callbacks
)
# Evaluate the model
loss, accuracy = model_res.evaluate(validation_generator1)
print(f"Validation Accuracy: {accuracy * 100:.2f}%")
# Save the trained model
model.save('er_30_epochs_resnet.keras')
print("Model saved as 'er_30_epochs_resnet.keras'.")
Epoch 1/30
Exception ignored in: <function AtomicFunction.__del__ at 0x16dabc700>
Traceback (most recent call last):
File "/Users/saeedesai/Library/Python/3.9/lib/python/site-packages/tensorflow/python/eager/polymorphic_function/atomic_function.py", line 286, in __del__
def __del__(self):
KeyboardInterrupt:
408/408 ━━━━━━━━━━━━━━━━━━━━ 323s 764ms/step - accuracy: 0.7352 - loss: 0.7075 - val_accuracy: 0.5737 - val_loss: 1.1315 - learning_rate: 5.0000e-05 Epoch 2/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 310s 761ms/step - accuracy: 0.7367 - loss: 0.7000 - val_accuracy: 0.5953 - val_loss: 1.0619 - learning_rate: 5.0000e-05 Epoch 3/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.7399 - loss: 0.6923 - val_accuracy: 0.5937 - val_loss: 1.0906 - learning_rate: 5.0000e-05 Epoch 4/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 309s 758ms/step - accuracy: 0.7493 - loss: 0.6703 - val_accuracy: 0.6026 - val_loss: 1.0547 - learning_rate: 5.0000e-05 Epoch 5/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 312s 766ms/step - accuracy: 0.7466 - loss: 0.6693 - val_accuracy: 0.5997 - val_loss: 1.0734 - learning_rate: 5.0000e-05 Epoch 6/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 315s 772ms/step - accuracy: 0.7537 - loss: 0.6609 - val_accuracy: 0.5991 - val_loss: 1.0767 - learning_rate: 5.0000e-05 Epoch 7/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 322s 789ms/step - accuracy: 0.7570 - loss: 0.6432 - val_accuracy: 0.6207 - val_loss: 1.0322 - learning_rate: 5.0000e-05 Epoch 8/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 313s 767ms/step - accuracy: 0.7586 - loss: 0.6394 - val_accuracy: 0.6244 - val_loss: 1.0172 - learning_rate: 5.0000e-05 Epoch 9/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.7619 - loss: 0.6345 - val_accuracy: 0.6260 - val_loss: 1.0169 - learning_rate: 5.0000e-05 Epoch 10/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 747ms/step - accuracy: 0.7671 - loss: 0.6183 - val_accuracy: 0.6293 - val_loss: 1.0037 - learning_rate: 5.0000e-05 Epoch 11/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 306s 749ms/step - accuracy: 0.7691 - loss: 0.6179 - val_accuracy: 0.6094 - val_loss: 1.0560 - learning_rate: 5.0000e-05 Epoch 12/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.7712 - loss: 0.6103 - val_accuracy: 0.6235 - val_loss: 1.0403 - learning_rate: 5.0000e-05 Epoch 13/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 747ms/step - accuracy: 0.7752 - loss: 0.6102 - val_accuracy: 0.6364 - val_loss: 1.0002 - learning_rate: 5.0000e-05 Epoch 14/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 306s 749ms/step - accuracy: 0.7754 - loss: 0.5997 - val_accuracy: 0.6372 - val_loss: 0.9900 - learning_rate: 5.0000e-05 Epoch 15/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 303s 744ms/step - accuracy: 0.7772 - loss: 0.5995 - val_accuracy: 0.6312 - val_loss: 1.0062 - learning_rate: 5.0000e-05 Epoch 16/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 300s 734ms/step - accuracy: 0.7812 - loss: 0.5849 - val_accuracy: 0.6198 - val_loss: 1.0912 - learning_rate: 5.0000e-05 Epoch 17/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 0s 717ms/step - accuracy: 0.7786 - loss: 0.5904 Epoch 17: ReduceLROnPlateau reducing learning rate to 9.999999747378752e-06. 408/408 ━━━━━━━━━━━━━━━━━━━━ 300s 734ms/step - accuracy: 0.7786 - loss: 0.5904 - val_accuracy: 0.6340 - val_loss: 1.0043 - learning_rate: 5.0000e-05 Epoch 18/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 1484s 4s/step - accuracy: 0.7913 - loss: 0.5616 - val_accuracy: 0.6520 - val_loss: 0.9578 - learning_rate: 1.0000e-05 Epoch 19/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 301s 738ms/step - accuracy: 0.7960 - loss: 0.5476 - val_accuracy: 0.6490 - val_loss: 0.9607 - learning_rate: 1.0000e-05 Epoch 20/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 302s 740ms/step - accuracy: 0.7926 - loss: 0.5513 - val_accuracy: 0.6540 - val_loss: 0.9610 - learning_rate: 1.0000e-05 Epoch 21/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 0s 729ms/step - accuracy: 0.7987 - loss: 0.5389 Epoch 21: ReduceLROnPlateau reducing learning rate to 1.9999999494757505e-06. 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 747ms/step - accuracy: 0.7987 - loss: 0.5389 - val_accuracy: 0.6527 - val_loss: 0.9686 - learning_rate: 1.0000e-05 Epoch 22/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.8007 - loss: 0.5328 - val_accuracy: 0.6562 - val_loss: 0.9608 - learning_rate: 2.0000e-06 Epoch 23/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 306s 751ms/step - accuracy: 0.7998 - loss: 0.5355 - val_accuracy: 0.6575 - val_loss: 0.9567 - learning_rate: 2.0000e-06 Epoch 24/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 306s 749ms/step - accuracy: 0.7969 - loss: 0.5399 - val_accuracy: 0.6574 - val_loss: 0.9577 - learning_rate: 2.0000e-06 Epoch 25/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.8023 - loss: 0.5241 - val_accuracy: 0.6583 - val_loss: 0.9564 - learning_rate: 2.0000e-06 Epoch 26/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 307s 753ms/step - accuracy: 0.8061 - loss: 0.5217 - val_accuracy: 0.6574 - val_loss: 0.9563 - learning_rate: 2.0000e-06 Epoch 27/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 307s 753ms/step - accuracy: 0.8025 - loss: 0.5289 - val_accuracy: 0.6574 - val_loss: 0.9558 - learning_rate: 2.0000e-06 Epoch 28/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 308s 754ms/step - accuracy: 0.8042 - loss: 0.5243 - val_accuracy: 0.6595 - val_loss: 0.9541 - learning_rate: 2.0000e-06 Epoch 29/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.8048 - loss: 0.5214 - val_accuracy: 0.6588 - val_loss: 0.9524 - learning_rate: 2.0000e-06 Epoch 30/30 408/408 ━━━━━━━━━━━━━━━━━━━━ 305s 748ms/step - accuracy: 0.8044 - loss: 0.5243 - val_accuracy: 0.6620 - val_loss: 0.9518 - learning_rate: 2.0000e-06 64/64 ━━━━━━━━━━━━━━━━━━━━ 7s 116ms/step - accuracy: 0.6615 - loss: 0.9572 Validation Accuracy: 66.20% Model saved as '30_epochs_resnet.keras'.
In [56]:
import matplotlib.pyplot as plt
# Retrieve metrics from the history object
train_loss = history.history['loss']
val_loss = history.history['val_loss']
train_accuracy = history.history['accuracy']
val_accuracy = history.history['val_accuracy']
epochs = range(1, len(train_loss) + 1)
# Plot Loss
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(epochs, train_loss, label='Training Loss', color='blue')
plt.plot(epochs, val_loss, label='Validation Loss', color='orange')
plt.title('Training and Validation Loss for RESNET50')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.grid(True)
# Plot Accuracy
plt.subplot(1, 2, 2)
plt.plot(epochs, train_accuracy, label='Training Accuracy', color='blue')
plt.plot(epochs, val_accuracy, label='Validation Accuracy', color='orange')
plt.title('Training and Validation Accuracy for RESNET50')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.grid(True)
# Show the plots
plt.tight_layout()
plt.show()
Transfer learning with MobileNetV2 as the base, adding custom fully connected layers, and compiling the model for 8-class image classification using categorical crossentropy and the Adam optimizer.¶
In [99]:
from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.layers import Dense, Dropout, GlobalAveragePooling2D, Flatten
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
conv_base = MobileNetV2(
weights='imagenet',
include_top=False,
input_shape=(48, 48, 3)
)
model_mob = Sequential()
model_mob.add(conv_base)
model_mob.add(Flatten())
model_mob.add(Dense(512,activation = 'relu'))
model_mob.add(Dense(256,activation = 'relu'))
model_mob.add(Dropout(0.45))
model_mob.add(Dense(8,activation='softmax'))
# Compile the model
model_mob.compile(
optimizer=Adam(learning_rate=0.00005),
loss='categorical_crossentropy',
metrics=['accuracy']
)
conv_base.summary()
/var/folders/dt/mx45884x0t9203dkrz1sq90r0000gn/T/ipykernel_6923/487814869.py:7: UserWarning: `input_shape` is undefined or non-square, or `rows` is not in [96, 128, 160, 192, 224]. Weights for input shape (224, 224) will be loaded as the default. conv_base = MobileNetV2(
Model: "mobilenetv2_1.00_224"
┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ Connected to ┃ ┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩ │ input_layer_29 │ (None, 48, 48, 3) │ 0 │ - │ │ (InputLayer) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ Conv1 (Conv2D) │ (None, 24, 24, │ 864 │ input_layer_29[0… │ │ │ 32) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ bn_Conv1 │ (None, 24, 24, │ 128 │ Conv1[0][0] │ │ (BatchNormalizatio… │ 32) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ Conv1_relu (ReLU) │ (None, 24, 24, │ 0 │ bn_Conv1[0][0] │ │ │ 32) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ expanded_conv_dept… │ (None, 24, 24, │ 288 │ Conv1_relu[0][0] │ │ (DepthwiseConv2D) │ 32) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ expanded_conv_dept… │ (None, 24, 24, │ 128 │ expanded_conv_de… │ │ (BatchNormalizatio… │ 32) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ expanded_conv_dept… │ (None, 24, 24, │ 0 │ expanded_conv_de… │ │ (ReLU) │ 32) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ expanded_conv_proj… │ (None, 24, 24, │ 512 │ expanded_conv_de… │ │ (Conv2D) │ 16) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ expanded_conv_proj… │ (None, 24, 24, │ 64 │ expanded_conv_pr… │ │ (BatchNormalizatio… │ 16) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_expand │ (None, 24, 24, │ 1,536 │ expanded_conv_pr… │ │ (Conv2D) │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_expand_BN │ (None, 24, 24, │ 384 │ block_1_expand[0… │ │ (BatchNormalizatio… │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_expand_relu │ (None, 24, 24, │ 0 │ block_1_expand_B… │ │ (ReLU) │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_pad │ (None, 25, 25, │ 0 │ block_1_expand_r… │ │ (ZeroPadding2D) │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_depthwise │ (None, 12, 12, │ 864 │ block_1_pad[0][0] │ │ (DepthwiseConv2D) │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_depthwise_… │ (None, 12, 12, │ 384 │ block_1_depthwis… │ │ (BatchNormalizatio… │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_depthwise_… │ (None, 12, 12, │ 0 │ block_1_depthwis… │ │ (ReLU) │ 96) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_project │ (None, 12, 12, │ 2,304 │ block_1_depthwis… │ │ (Conv2D) │ 24) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_1_project_BN │ (None, 12, 12, │ 96 │ block_1_project[… │ │ (BatchNormalizatio… │ 24) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_expand │ (None, 12, 12, │ 3,456 │ block_1_project_… │ │ (Conv2D) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_expand_BN │ (None, 12, 12, │ 576 │ block_2_expand[0… │ │ (BatchNormalizatio… │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_expand_relu │ (None, 12, 12, │ 0 │ block_2_expand_B… │ │ (ReLU) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_depthwise │ (None, 12, 12, │ 1,296 │ block_2_expand_r… │ │ (DepthwiseConv2D) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_depthwise_… │ (None, 12, 12, │ 576 │ block_2_depthwis… │ │ (BatchNormalizatio… │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_depthwise_… │ (None, 12, 12, │ 0 │ block_2_depthwis… │ │ (ReLU) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_project │ (None, 12, 12, │ 3,456 │ block_2_depthwis… │ │ (Conv2D) │ 24) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_project_BN │ (None, 12, 12, │ 96 │ block_2_project[… │ │ (BatchNormalizatio… │ 24) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_2_add (Add) │ (None, 12, 12, │ 0 │ block_1_project_… │ │ │ 24) │ │ block_2_project_… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_expand │ (None, 12, 12, │ 3,456 │ block_2_add[0][0] │ │ (Conv2D) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_expand_BN │ (None, 12, 12, │ 576 │ block_3_expand[0… │ │ (BatchNormalizatio… │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_expand_relu │ (None, 12, 12, │ 0 │ block_3_expand_B… │ │ (ReLU) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_pad │ (None, 13, 13, │ 0 │ block_3_expand_r… │ │ (ZeroPadding2D) │ 144) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_depthwise │ (None, 6, 6, 144) │ 1,296 │ block_3_pad[0][0] │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_depthwise_… │ (None, 6, 6, 144) │ 576 │ block_3_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_depthwise_… │ (None, 6, 6, 144) │ 0 │ block_3_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_project │ (None, 6, 6, 32) │ 4,608 │ block_3_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_3_project_BN │ (None, 6, 6, 32) │ 128 │ block_3_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_expand │ (None, 6, 6, 192) │ 6,144 │ block_3_project_… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_expand_BN │ (None, 6, 6, 192) │ 768 │ block_4_expand[0… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_expand_relu │ (None, 6, 6, 192) │ 0 │ block_4_expand_B… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_depthwise │ (None, 6, 6, 192) │ 1,728 │ block_4_expand_r… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_depthwise_… │ (None, 6, 6, 192) │ 768 │ block_4_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_depthwise_… │ (None, 6, 6, 192) │ 0 │ block_4_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_project │ (None, 6, 6, 32) │ 6,144 │ block_4_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_project_BN │ (None, 6, 6, 32) │ 128 │ block_4_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_4_add (Add) │ (None, 6, 6, 32) │ 0 │ block_3_project_… │ │ │ │ │ block_4_project_… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_expand │ (None, 6, 6, 192) │ 6,144 │ block_4_add[0][0] │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_expand_BN │ (None, 6, 6, 192) │ 768 │ block_5_expand[0… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_expand_relu │ (None, 6, 6, 192) │ 0 │ block_5_expand_B… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_depthwise │ (None, 6, 6, 192) │ 1,728 │ block_5_expand_r… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_depthwise_… │ (None, 6, 6, 192) │ 768 │ block_5_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_depthwise_… │ (None, 6, 6, 192) │ 0 │ block_5_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_project │ (None, 6, 6, 32) │ 6,144 │ block_5_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_project_BN │ (None, 6, 6, 32) │ 128 │ block_5_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_5_add (Add) │ (None, 6, 6, 32) │ 0 │ block_4_add[0][0… │ │ │ │ │ block_5_project_… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_expand │ (None, 6, 6, 192) │ 6,144 │ block_5_add[0][0] │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_expand_BN │ (None, 6, 6, 192) │ 768 │ block_6_expand[0… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_expand_relu │ (None, 6, 6, 192) │ 0 │ block_6_expand_B… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_pad │ (None, 7, 7, 192) │ 0 │ block_6_expand_r… │ │ (ZeroPadding2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_depthwise │ (None, 3, 3, 192) │ 1,728 │ block_6_pad[0][0] │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_depthwise_… │ (None, 3, 3, 192) │ 768 │ block_6_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_depthwise_… │ (None, 3, 3, 192) │ 0 │ block_6_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_project │ (None, 3, 3, 64) │ 12,288 │ block_6_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_6_project_BN │ (None, 3, 3, 64) │ 256 │ block_6_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_expand │ (None, 3, 3, 384) │ 24,576 │ block_6_project_… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_expand_BN │ (None, 3, 3, 384) │ 1,536 │ block_7_expand[0… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_expand_relu │ (None, 3, 3, 384) │ 0 │ block_7_expand_B… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_depthwise │ (None, 3, 3, 384) │ 3,456 │ block_7_expand_r… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_depthwise_… │ (None, 3, 3, 384) │ 1,536 │ block_7_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_depthwise_… │ (None, 3, 3, 384) │ 0 │ block_7_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_project │ (None, 3, 3, 64) │ 24,576 │ block_7_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_project_BN │ (None, 3, 3, 64) │ 256 │ block_7_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_7_add (Add) │ (None, 3, 3, 64) │ 0 │ block_6_project_… │ │ │ │ │ block_7_project_… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_expand │ (None, 3, 3, 384) │ 24,576 │ block_7_add[0][0] │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_expand_BN │ (None, 3, 3, 384) │ 1,536 │ block_8_expand[0… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_expand_relu │ (None, 3, 3, 384) │ 0 │ block_8_expand_B… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_depthwise │ (None, 3, 3, 384) │ 3,456 │ block_8_expand_r… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_depthwise_… │ (None, 3, 3, 384) │ 1,536 │ block_8_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_depthwise_… │ (None, 3, 3, 384) │ 0 │ block_8_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_project │ (None, 3, 3, 64) │ 24,576 │ block_8_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_project_BN │ (None, 3, 3, 64) │ 256 │ block_8_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_8_add (Add) │ (None, 3, 3, 64) │ 0 │ block_7_add[0][0… │ │ │ │ │ block_8_project_… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_expand │ (None, 3, 3, 384) │ 24,576 │ block_8_add[0][0] │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_expand_BN │ (None, 3, 3, 384) │ 1,536 │ block_9_expand[0… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_expand_relu │ (None, 3, 3, 384) │ 0 │ block_9_expand_B… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_depthwise │ (None, 3, 3, 384) │ 3,456 │ block_9_expand_r… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_depthwise_… │ (None, 3, 3, 384) │ 1,536 │ block_9_depthwis… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_depthwise_… │ (None, 3, 3, 384) │ 0 │ block_9_depthwis… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_project │ (None, 3, 3, 64) │ 24,576 │ block_9_depthwis… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_project_BN │ (None, 3, 3, 64) │ 256 │ block_9_project[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_9_add (Add) │ (None, 3, 3, 64) │ 0 │ block_8_add[0][0… │ │ │ │ │ block_9_project_… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_expand │ (None, 3, 3, 384) │ 24,576 │ block_9_add[0][0] │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_expand_BN │ (None, 3, 3, 384) │ 1,536 │ block_10_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_expand_re… │ (None, 3, 3, 384) │ 0 │ block_10_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_depthwise │ (None, 3, 3, 384) │ 3,456 │ block_10_expand_… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_depthwise… │ (None, 3, 3, 384) │ 1,536 │ block_10_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_depthwise… │ (None, 3, 3, 384) │ 0 │ block_10_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_project │ (None, 3, 3, 96) │ 36,864 │ block_10_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_10_project_BN │ (None, 3, 3, 96) │ 384 │ block_10_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_expand │ (None, 3, 3, 576) │ 55,296 │ block_10_project… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_expand_BN │ (None, 3, 3, 576) │ 2,304 │ block_11_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_expand_re… │ (None, 3, 3, 576) │ 0 │ block_11_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_depthwise │ (None, 3, 3, 576) │ 5,184 │ block_11_expand_… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_depthwise… │ (None, 3, 3, 576) │ 2,304 │ block_11_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_depthwise… │ (None, 3, 3, 576) │ 0 │ block_11_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_project │ (None, 3, 3, 96) │ 55,296 │ block_11_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_project_BN │ (None, 3, 3, 96) │ 384 │ block_11_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_11_add (Add) │ (None, 3, 3, 96) │ 0 │ block_10_project… │ │ │ │ │ block_11_project… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_expand │ (None, 3, 3, 576) │ 55,296 │ block_11_add[0][… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_expand_BN │ (None, 3, 3, 576) │ 2,304 │ block_12_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_expand_re… │ (None, 3, 3, 576) │ 0 │ block_12_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_depthwise │ (None, 3, 3, 576) │ 5,184 │ block_12_expand_… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_depthwise… │ (None, 3, 3, 576) │ 2,304 │ block_12_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_depthwise… │ (None, 3, 3, 576) │ 0 │ block_12_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_project │ (None, 3, 3, 96) │ 55,296 │ block_12_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_project_BN │ (None, 3, 3, 96) │ 384 │ block_12_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_12_add (Add) │ (None, 3, 3, 96) │ 0 │ block_11_add[0][… │ │ │ │ │ block_12_project… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_expand │ (None, 3, 3, 576) │ 55,296 │ block_12_add[0][… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_expand_BN │ (None, 3, 3, 576) │ 2,304 │ block_13_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_expand_re… │ (None, 3, 3, 576) │ 0 │ block_13_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_pad │ (None, 5, 5, 576) │ 0 │ block_13_expand_… │ │ (ZeroPadding2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_depthwise │ (None, 2, 2, 576) │ 5,184 │ block_13_pad[0][… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_depthwise… │ (None, 2, 2, 576) │ 2,304 │ block_13_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_depthwise… │ (None, 2, 2, 576) │ 0 │ block_13_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_project │ (None, 2, 2, 160) │ 92,160 │ block_13_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_13_project_BN │ (None, 2, 2, 160) │ 640 │ block_13_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_expand │ (None, 2, 2, 960) │ 153,600 │ block_13_project… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_expand_BN │ (None, 2, 2, 960) │ 3,840 │ block_14_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_expand_re… │ (None, 2, 2, 960) │ 0 │ block_14_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_depthwise │ (None, 2, 2, 960) │ 8,640 │ block_14_expand_… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_depthwise… │ (None, 2, 2, 960) │ 3,840 │ block_14_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_depthwise… │ (None, 2, 2, 960) │ 0 │ block_14_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_project │ (None, 2, 2, 160) │ 153,600 │ block_14_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_project_BN │ (None, 2, 2, 160) │ 640 │ block_14_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_14_add (Add) │ (None, 2, 2, 160) │ 0 │ block_13_project… │ │ │ │ │ block_14_project… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_expand │ (None, 2, 2, 960) │ 153,600 │ block_14_add[0][… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_expand_BN │ (None, 2, 2, 960) │ 3,840 │ block_15_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_expand_re… │ (None, 2, 2, 960) │ 0 │ block_15_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_depthwise │ (None, 2, 2, 960) │ 8,640 │ block_15_expand_… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_depthwise… │ (None, 2, 2, 960) │ 3,840 │ block_15_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_depthwise… │ (None, 2, 2, 960) │ 0 │ block_15_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_project │ (None, 2, 2, 160) │ 153,600 │ block_15_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_project_BN │ (None, 2, 2, 160) │ 640 │ block_15_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_15_add (Add) │ (None, 2, 2, 160) │ 0 │ block_14_add[0][… │ │ │ │ │ block_15_project… │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_expand │ (None, 2, 2, 960) │ 153,600 │ block_15_add[0][… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_expand_BN │ (None, 2, 2, 960) │ 3,840 │ block_16_expand[… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_expand_re… │ (None, 2, 2, 960) │ 0 │ block_16_expand_… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_depthwise │ (None, 2, 2, 960) │ 8,640 │ block_16_expand_… │ │ (DepthwiseConv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_depthwise… │ (None, 2, 2, 960) │ 3,840 │ block_16_depthwi… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_depthwise… │ (None, 2, 2, 960) │ 0 │ block_16_depthwi… │ │ (ReLU) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_project │ (None, 2, 2, 320) │ 307,200 │ block_16_depthwi… │ │ (Conv2D) │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ block_16_project_BN │ (None, 2, 2, 320) │ 1,280 │ block_16_project… │ │ (BatchNormalizatio… │ │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ Conv_1 (Conv2D) │ (None, 2, 2, │ 409,600 │ block_16_project… │ │ │ 1280) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ Conv_1_bn │ (None, 2, 2, │ 5,120 │ Conv_1[0][0] │ │ (BatchNormalizatio… │ 1280) │ │ │ ├─────────────────────┼───────────────────┼────────────┼───────────────────┤ │ out_relu (ReLU) │ (None, 2, 2, │ 0 │ Conv_1_bn[0][0] │ │ │ 1280) │ │ │ └─────────────────────┴───────────────────┴────────────┴───────────────────┘
Total params: 2,257,984 (8.61 MB)
Trainable params: 2,223,872 (8.48 MB)
Non-trainable params: 34,112 (133.25 KB)
In [ ]:
# Define callbacks
callbacks = [
EarlyStopping(
monitor='val_loss',
patience=5,
restore_best_weights=True
),
ReduceLROnPlateau(
monitor='val_loss',
factor=0.2,
patience=3,
verbose=1
)
]
# Training the Model
history = model_mob.fit(
train_generator2,
epochs=100,
validation_data=validation_generator1,
callbacks = callbacks
)
# Evaluate the model
loss, accuracy = model_mob.evaluate(validation_generator1)
print(f"Validation Accuracy: {accuracy * 100:.2f}%")
# Save the trained model
model.save('100epochs_mobilenet.keras')
print("Model saved as '100epochs_mobilenet.keras'.")
In [66]:
import matplotlib.pyplot as plt
# Retrieve metrics from the history object
train_loss = history.history['loss']
val_loss = history.history['val_loss']
train_accuracy = history.history['accuracy']
val_accuracy = history.history['val_accuracy']
epochs = range(1, len(train_loss) + 1)
# Plot Loss
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(epochs, train_loss, label='Training Loss', color='blue')
plt.plot(epochs, val_loss, label='Validation Loss', color='orange')
plt.title('Training and Validation Loss for MobileNet')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.grid(True)
# Plot Accuracy
plt.subplot(1, 2, 2)
plt.plot(epochs, train_accuracy, label='Training Accuracy', color='blue')
plt.plot(epochs, val_accuracy, label='Validation Accuracy', color='orange')
plt.title('Training and Validation Accuracy for MobileNet')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.grid(True)
# Show the plots
plt.tight_layout()
plt.show()
Using K-FOLD Cross Validation¶
In [77]:
import os
import shutil
combined_dir = "/Users/saeedesai/Desktop/My codes/DATA602_FINAL_PROJECT/train_val"
# Create the combined directory
os.makedirs(combined_dir, exist_ok=True)
# Function to move files
def move_files(source_dir, target_dir):
for class_name in os.listdir(source_dir):
src_class_dir = os.path.join(source_dir, class_name)
target_class_dir = os.path.join(target_dir, class_name)
os.makedirs(target_class_dir, exist_ok=True)
for file_name in os.listdir(src_class_dir):
shutil.move(os.path.join(src_class_dir, file_name), target_class_dir)
# Move train and val data into combined_data
move_files(train_path, combined_dir)
move_files(val_path, combined_dir)
print("Train and Validation folders combined.")
Train and Validation folders combined.
5-fold cross-validation on an image classification task using a custom ConvNet2 architecture in PyTorch. The model trains for 25 epochs per fold, evaluates on validation data, and computes fold-wise metrics (loss and accuracy), followed by averaging the results.¶
In [78]:
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader, Subset, ConcatDataset
from torchvision import datasets, transforms
from sklearn.model_selection import KFold
import numpy as np
# Define your ConvNet2 model
class ConvNet2(nn.Module):
def __init__(self):
super(ConvNet2, self).__init__()
self.c1 = nn.Conv2d(1, 32, 3, padding=1)
self.c2 = nn.Conv2d(32, 64, 3, padding=1)
self.c3 = nn.Conv2d(64, 128, 5, padding=2)
self.c4 = nn.Conv2d(128, 512, 3, padding=1)
self.c5 = nn.Conv2d(512, 512, 3, padding=1)
self.bn1 = nn.BatchNorm2d(64)
self.bn2 = nn.BatchNorm2d(128)
self.bn3 = nn.BatchNorm2d(512)
self.bn4 = nn.BatchNorm2d(512)
self.bn5 = nn.BatchNorm1d(256)
self.bn6 = nn.BatchNorm1d(512)
self.l1 = nn.Linear(512 * 3 * 3, 256)
self.l2 = nn.Linear(256, 512)
self.l3 = nn.Linear(512, 8)
self.relu = nn.ReLU()
self.pool = nn.MaxPool2d(2, 2)
self.dropout = nn.Dropout(0.25)
def forward(self, x):
x = F.relu(self.c1(x))
x = F.relu(self.c2(x))
x = self.dropout(self.pool(self.bn1(x)))
x = F.relu(self.c3(x))
x = self.dropout(self.pool(self.bn2(x)))
x = F.relu(self.c4(x))
x = self.dropout(self.pool(self.bn3(x)))
x = F.relu(self.c5(x))
x = self.dropout(self.pool(self.bn4(x)))
x = x.view(-1, 512 * 3 * 3)
x = self.bn5(F.relu(self.l1(x)))
x = self.bn6(F.relu(self.l2(x)))
x = self.l3(x)
return x
# Define your dataset and transforms
transform = transforms.Compose([
transforms.Grayscale(num_output_channels=1),
transforms.Resize((48, 48)),
transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,))
])
dataset = datasets.ImageFolder(combined_dir, transform=transform)
# Initialize KFold
kfold = KFold(n_splits=5, shuffle=True, random_state=42)
# Set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Results storage
fold_results = []
for fold, (train_idx, val_idx) in enumerate(kfold.split(dataset)):
print(f"Processing Fold {fold + 1}")
# Subset the dataset
train_subset = Subset(dataset, train_idx)
val_subset = Subset(dataset, val_idx)
# Data loaders
train_loader = DataLoader(train_subset, batch_size=128, shuffle=True)
val_loader = DataLoader(val_subset, batch_size=128, shuffle=False)
# Initialize the model, loss, and optimizer
model = ConvNet2().to(device)
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.0001)
# Train the model
for epoch in range(25): # Adjust epochs as needed
model.train()
running_loss = 0.0
for inputs, labels in train_loader:
inputs, labels = inputs.to(device), labels.to(device)
# Zero the parameter gradients
optimizer.zero_grad()
# Forward + backward + optimize
outputs = model(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
print(f"Fold {fold + 1}, Epoch {epoch + 1}, Loss: {running_loss / len(train_loader)}")
# Evaluate the model
model.eval()
val_loss = 0.0
correct = 0
total = 0
with torch.no_grad():
for inputs, labels in val_loader:
inputs, labels = inputs.to(device), labels.to(device)
outputs = model(inputs)
loss = criterion(outputs, labels)
val_loss += loss.item()
# Calculate accuracy
_, predicted = torch.max(outputs, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
val_accuracy = 100 * correct / total
print(f"Fold {fold + 1}, Validation Loss: {val_loss / len(val_loader)}, Accuracy: {val_accuracy:.2f}%")
# Store results
fold_results.append({"fold": fold + 1, "val_loss": val_loss / len(val_loader), "val_accuracy": val_accuracy})
# Calculate average accuracy and loss
avg_val_loss = np.mean([result["val_loss"] for result in fold_results])
avg_val_accuracy = np.mean([result["val_accuracy"] for result in fold_results])
print(f"\nAverage Validation Loss: {avg_val_loss:.4f}")
print(f"Average Validation Accuracy: {avg_val_accuracy:.2f}%")
model.save('kfold_trained.h5')
Processing Fold 1 Fold 1, Epoch 1, Loss: 1.7252393689582017 Fold 1, Epoch 2, Loss: 1.4470895591790114 Fold 1, Epoch 3, Loss: 1.3089385430018108 Fold 1, Epoch 4, Loss: 1.2194503288928087 Fold 1, Epoch 5, Loss: 1.143337916552536 Fold 1, Epoch 6, Loss: 1.0692618947203567 Fold 1, Epoch 7, Loss: 1.0111743185093733 Fold 1, Epoch 8, Loss: 0.9578879793000415 Fold 1, Epoch 9, Loss: 0.8997712542371052 Fold 1, Epoch 10, Loss: 0.8446610073248545 Fold 1, Epoch 11, Loss: 0.7806052284512094 Fold 1, Epoch 12, Loss: 0.7240517599795891 Fold 1, Epoch 13, Loss: 0.661685368394464 Fold 1, Epoch 14, Loss: 0.6060197180121895 Fold 1, Epoch 15, Loss: 0.550810551982585 Fold 1, Epoch 16, Loss: 0.4952950209863787 Fold 1, Epoch 17, Loss: 0.44522726911354843 Fold 1, Epoch 18, Loss: 0.3984773551545492 Fold 1, Epoch 19, Loss: 0.36059671610109206 Fold 1, Epoch 20, Loss: 0.3226463681435197 Fold 1, Epoch 21, Loss: 0.2880730426044968 Fold 1, Epoch 22, Loss: 0.2644820329619617 Fold 1, Epoch 23, Loss: 0.24483866165808546 Fold 1, Epoch 24, Loss: 0.2208995041021002 Fold 1, Epoch 25, Loss: 0.2024024812792375 Fold 1, Validation Loss: 1.4885614821987767, Accuracy: 63.10% Processing Fold 2 Fold 2, Epoch 1, Loss: 1.7440756045705903 Fold 2, Epoch 2, Loss: 1.4600648584404612 Fold 2, Epoch 3, Loss: 1.3199050353794564 Fold 2, Epoch 4, Loss: 1.2218693928505346 Fold 2, Epoch 5, Loss: 1.143309472295327 Fold 2, Epoch 6, Loss: 1.0708199274249193 Fold 2, Epoch 7, Loss: 1.0086543700559352 Fold 2, Epoch 8, Loss: 0.9514133217858105 Fold 2, Epoch 9, Loss: 0.8947379918602424 Fold 2, Epoch 10, Loss: 0.8326893373718106 Fold 2, Epoch 11, Loss: 0.7724736046984912 Fold 2, Epoch 12, Loss: 0.7142949491981568 Fold 2, Epoch 13, Loss: 0.6584494550053667 Fold 2, Epoch 14, Loss: 0.6007876868655042 Fold 2, Epoch 15, Loss: 0.5393373373805023 Fold 2, Epoch 16, Loss: 0.4980279965613916 Fold 2, Epoch 17, Loss: 0.43559779917321556 Fold 2, Epoch 18, Loss: 0.39359775143183345 Fold 2, Epoch 19, Loss: 0.35415520107116155 Fold 2, Epoch 20, Loss: 0.3270416551125728 Fold 2, Epoch 21, Loss: 0.2906980611444489 Fold 2, Epoch 22, Loss: 0.2653853097339956 Fold 2, Epoch 23, Loss: 0.24009291821621298 Fold 2, Epoch 24, Loss: 0.21308980536897007 Fold 2, Epoch 25, Loss: 0.2026454862661478 Fold 2, Validation Loss: 1.4432584422249948, Accuracy: 64.50% Processing Fold 3 Fold 3, Epoch 1, Loss: 1.7587828229113323 Fold 3, Epoch 2, Loss: 1.4675351691439869 Fold 3, Epoch 3, Loss: 1.3344553553961156 Fold 3, Epoch 4, Loss: 1.2294795789369724 Fold 3, Epoch 5, Loss: 1.1506117403022642 Fold 3, Epoch 6, Loss: 1.0831702448488252 Fold 3, Epoch 7, Loss: 1.0228096445401509 Fold 3, Epoch 8, Loss: 0.961421015301371 Fold 3, Epoch 9, Loss: 0.91081839024536 Fold 3, Epoch 10, Loss: 0.8533999779360081 Fold 3, Epoch 11, Loss: 0.7880894253409005 Fold 3, Epoch 12, Loss: 0.7361663919154221 Fold 3, Epoch 13, Loss: 0.6726869235678417 Fold 3, Epoch 14, Loss: 0.6160195669749888 Fold 3, Epoch 15, Loss: 0.5623159243808529 Fold 3, Epoch 16, Loss: 0.5074979946380709 Fold 3, Epoch 17, Loss: 0.4643653881985967 Fold 3, Epoch 18, Loss: 0.41450292423246354 Fold 3, Epoch 19, Loss: 0.3645596200736558 Fold 3, Epoch 20, Loss: 0.33104023579659503 Fold 3, Epoch 21, Loss: 0.30261203513397433 Fold 3, Epoch 22, Loss: 0.2669750548354009 Fold 3, Epoch 23, Loss: 0.2528272757564134 Fold 3, Epoch 24, Loss: 0.23316227705255757 Fold 3, Epoch 25, Loss: 0.20870032621835305 Fold 3, Validation Loss: 1.362069294817986, Accuracy: 64.24% Processing Fold 4 Fold 4, Epoch 1, Loss: 1.770509162084843 Fold 4, Epoch 2, Loss: 1.4796966168938614 Fold 4, Epoch 3, Loss: 1.345336428502711 Fold 4, Epoch 4, Loss: 1.243209176916417 Fold 4, Epoch 5, Loss: 1.1642935435946395 Fold 4, Epoch 6, Loss: 1.0889836855050994 Fold 4, Epoch 7, Loss: 1.0321691043493224 Fold 4, Epoch 8, Loss: 0.9727674003054456 Fold 4, Epoch 9, Loss: 0.9150140583515167 Fold 4, Epoch 10, Loss: 0.8524204209083464 Fold 4, Epoch 11, Loss: 0.7920682309119682 Fold 4, Epoch 12, Loss: 0.7413502805601291 Fold 4, Epoch 13, Loss: 0.6773976076666902 Fold 4, Epoch 14, Loss: 0.6158980161436205 Fold 4, Epoch 15, Loss: 0.5587832378904994 Fold 4, Epoch 16, Loss: 0.50539559707409 Fold 4, Epoch 17, Loss: 0.4546569799020038 Fold 4, Epoch 18, Loss: 0.40536314224809167 Fold 4, Epoch 19, Loss: 0.3657083676719084 Fold 4, Epoch 20, Loss: 0.3306482925526495 Fold 4, Epoch 21, Loss: 0.30065419882293637 Fold 4, Epoch 22, Loss: 0.2613434846081385 Fold 4, Epoch 23, Loss: 0.24712291217552937 Fold 4, Epoch 24, Loss: 0.22436587918337766 Fold 4, Epoch 25, Loss: 0.20657803424126733 Fold 4, Validation Loss: 1.4303422389972595, Accuracy: 64.37% Processing Fold 5 Fold 5, Epoch 1, Loss: 1.7484565944206425 Fold 5, Epoch 2, Loss: 1.4712030160717848 Fold 5, Epoch 3, Loss: 1.3394548151551224 Fold 5, Epoch 4, Loss: 1.2347938297240715 Fold 5, Epoch 5, Loss: 1.155677107291493 Fold 5, Epoch 6, Loss: 1.0860509569567394 Fold 5, Epoch 7, Loss: 1.0297921847521774 Fold 5, Epoch 8, Loss: 0.9660251176938778 Fold 5, Epoch 9, Loss: 0.9128724033755016 Fold 5, Epoch 10, Loss: 0.8509719921805994 Fold 5, Epoch 11, Loss: 0.7899384818425993 Fold 5, Epoch 12, Loss: 0.7297504730825501 Fold 5, Epoch 13, Loss: 0.6709904482936472 Fold 5, Epoch 14, Loss: 0.6106691395606452 Fold 5, Epoch 15, Loss: 0.5582649305583985 Fold 5, Epoch 16, Loss: 0.5042594945043083 Fold 5, Epoch 17, Loss: 0.455676378757973 Fold 5, Epoch 18, Loss: 0.4020938108849332 Fold 5, Epoch 19, Loss: 0.3662267495703891 Fold 5, Epoch 20, Loss: 0.3340833053235116 Fold 5, Epoch 21, Loss: 0.2936915639575904 Fold 5, Epoch 22, Loss: 0.26103231989271275 Fold 5, Epoch 23, Loss: 0.24539468134564113 Fold 5, Epoch 24, Loss: 0.2272465262047159 Fold 5, Epoch 25, Loss: 0.20695853421116264 Fold 5, Validation Loss: 1.3519067276389367, Accuracy: 64.54% Average Validation Loss: 1.4152 Average Validation Accuracy: 64.15%
--------------------------------------------------------------------------- AttributeError Traceback (most recent call last) Cell In[78], line 133 130 print(f"\nAverage Validation Loss: {avg_val_loss:.4f}") 131 print(f"Average Validation Accuracy: {avg_val_accuracy:.2f}%") --> 133 model.save('kfold_trained.h5') File ~/Library/Python/3.9/lib/python/site-packages/torch/nn/modules/module.py:1729, in Module.__getattr__(self, name) 1727 if name in modules: 1728 return modules[name] -> 1729 raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'") AttributeError: 'ConvNet2' object has no attribute 'save'
In [79]:
from torchvision.datasets import ImageFolder
from torchvision import transforms
from torch.utils.data import DataLoader, Subset
from sklearn.model_selection import KFold
# Define transforms for VGG16 (ImageNet mean and std normalization)
data_transforms = transforms.Compose([
transforms.Resize((48, 48)), # Resize images to 48x48
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) # ImageNet stats
])
# Load the dataset
dataset = ImageFolder(root=combined_dir, transform=data_transforms)
print(f"Total samples in combined dataset: {len(dataset)}")
Total samples in combined dataset: 39336
In [80]:
import torch.nn as nn
from torchvision.models import vgg16
class VGG16Model(nn.Module):
def __init__(self, num_classes=8):
super(VGG16Model, self).__init__()
# Load pre-trained VGG16 model
self.vgg = vgg16(pretrained=True)
# Modify the classifier for 8 classes
self.vgg.classifier[6] = nn.Linear(4096, num_classes)
def forward(self, x):
return self.vgg(x)
5-fold cross-validation for training a custom VGG16-based model. The dataset is split into training and validation sets for each fold, and a training loop runs for 10 epochs per fold. The model uses the Adam optimizer and cross-entropy loss.¶
In [84]:
from sklearn.model_selection import KFold
import torch
from torch.optim import Adam
# Define hyperparameters
k_folds = 5
batch_size = 128
learning_rate = 0.0001
num_epochs = 10
# Use KFold from sklearn
kfold = KFold(n_splits=k_folds, shuffle=True, random_state=42)
# Cross-validation loop
for fold, (train_idx, val_idx) in enumerate(kfold.split(dataset)):
print(f"Fold {fold + 1}/{k_folds}")
# Split dataset into training and validation subsets
train_subset = Subset(dataset, train_idx)
val_subset = Subset(dataset, val_idx)
# Create data loaders
train_loader = DataLoader(train_subset, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(val_subset, batch_size=batch_size, shuffle=False)
# Initialize model, optimizer, and loss function
model = VGG16Model(num_classes=8) # Move to GPU if available
optimizer = Adam(model.parameters(), lr=learning_rate)
criterion = nn.CrossEntropyLoss()
# Training and validation loop
for epoch in range(num_epochs):
print(f"Epoch {epoch + 1}/{num_epochs}")
# Training phase
model.train()
train_loss, train_acc = 0.0, 0.0
for inputs, labels in train_loader:
inputs, labels = inputs, labels
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
train_loss += loss.item()
train_acc += (outputs.argmax(dim=1) == labels).float().mean().item()
# Validation phase
model.eval()
val_loss, val_acc = 0.0, 0.0
with torch.no_grad():
for inputs, labels in val_loader:
inputs, labels = inputs, labels
outputs = model(inputs)
loss = criterion(outputs, labels)
val_loss += loss.item()
val_acc += (outputs.argmax(dim=1) == labels).float().mean().item()
print(f"Train Loss: {train_loss / len(train_loader):.4f}, Train Acc: {train_acc / len(train_loader):.4f}")
print(f"Val Loss: {val_loss / len(val_loader):.4f}, Val Acc: {val_acc / len(val_loader):.4f}")
# Save the model for the current fold
torch.save(model.state_dict(), f"vgg16_fold_{fold + 1}.pth")
print(f"Model for fold {fold + 1} saved.")
Fold 1/5
/Users/saeedesai/Library/Python/3.9/lib/python/site-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead. warnings.warn( /Users/saeedesai/Library/Python/3.9/lib/python/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=VGG16_Weights.IMAGENET1K_V1`. You can also use `weights=VGG16_Weights.DEFAULT` to get the most up-to-date weights. warnings.warn(msg)
Epoch 1/10 Train Loss: 1.3531, Train Acc: 0.4761 Val Loss: 1.1247, Val Acc: 0.5744 Epoch 2/10 Train Loss: 1.0247, Train Acc: 0.6150 Val Loss: 1.0360, Val Acc: 0.6102 Epoch 3/10 Train Loss: 0.8338, Train Acc: 0.6922 Val Loss: 0.9957, Val Acc: 0.6471 Epoch 4/10 Train Loss: 0.6508, Train Acc: 0.7651 Val Loss: 1.0804, Val Acc: 0.6424 Epoch 5/10 Train Loss: 0.4668, Train Acc: 0.8351 Val Loss: 1.0971, Val Acc: 0.6521 Epoch 6/10 Train Loss: 0.3347, Train Acc: 0.8831 Val Loss: 1.2263, Val Acc: 0.6565 Epoch 7/10 Train Loss: 0.2366, Train Acc: 0.9206 Val Loss: 1.3543, Val Acc: 0.6575 Epoch 8/10 Train Loss: 0.1761, Train Acc: 0.9394 Val Loss: 1.5588, Val Acc: 0.6539 Epoch 9/10 Train Loss: 0.1287, Train Acc: 0.9574 Val Loss: 1.6328, Val Acc: 0.6543 Epoch 10/10 Train Loss: 0.1017, Train Acc: 0.9665 Val Loss: 1.7108, Val Acc: 0.6517 Model for fold 1 saved. Fold 2/5 Epoch 1/10 Train Loss: 1.3505, Train Acc: 0.4743 Val Loss: 1.1151, Val Acc: 0.5754 Epoch 2/10 Train Loss: 1.0167, Train Acc: 0.6170 Val Loss: 1.0101, Val Acc: 0.6224 Epoch 3/10 Train Loss: 0.8392, Train Acc: 0.6917 Val Loss: 1.0881, Val Acc: 0.6173 Epoch 4/10 Train Loss: 0.6278, Train Acc: 0.7751 Val Loss: 1.0144, Val Acc: 0.6442 Epoch 5/10 Train Loss: 0.4455, Train Acc: 0.8423 Val Loss: 1.1693, Val Acc: 0.6613 Epoch 6/10 Train Loss: 0.3176, Train Acc: 0.8911 Val Loss: 1.2857, Val Acc: 0.6478 Epoch 7/10 Train Loss: 0.2190, Train Acc: 0.9255 Val Loss: 1.4140, Val Acc: 0.6441 Epoch 8/10 Train Loss: 0.1599, Train Acc: 0.9461 Val Loss: 1.6416, Val Acc: 0.6390 Epoch 9/10 Train Loss: 0.1391, Train Acc: 0.9534 Val Loss: 1.5782, Val Acc: 0.6629 Epoch 10/10 Train Loss: 0.1014, Train Acc: 0.9659 Val Loss: 1.6637, Val Acc: 0.6565 Model for fold 2 saved. Fold 3/5 Epoch 1/10 Train Loss: 1.3644, Train Acc: 0.4721 Val Loss: 1.1114, Val Acc: 0.5647 Epoch 2/10 Train Loss: 1.0278, Train Acc: 0.6124 Val Loss: 1.0336, Val Acc: 0.6067 Epoch 3/10 Train Loss: 0.8387, Train Acc: 0.6911 Val Loss: 0.9675, Val Acc: 0.6433 Epoch 4/10 Train Loss: 0.6713, Train Acc: 0.7571 Val Loss: 1.0043, Val Acc: 0.6441 Epoch 5/10 Train Loss: 0.4845, Train Acc: 0.8315 Val Loss: 1.1654, Val Acc: 0.6339 Epoch 6/10 Train Loss: 0.3435, Train Acc: 0.8821 Val Loss: 1.2129, Val Acc: 0.6614 Epoch 7/10 Train Loss: 0.2457, Train Acc: 0.9163 Val Loss: 1.3360, Val Acc: 0.6527 Epoch 8/10 Train Loss: 0.1838, Train Acc: 0.9365 Val Loss: 1.4569, Val Acc: 0.6623 Epoch 9/10 Train Loss: 0.1312, Train Acc: 0.9570 Val Loss: 1.6031, Val Acc: 0.6441 Epoch 10/10 Train Loss: 0.1143, Train Acc: 0.9621 Val Loss: 1.9881, Val Acc: 0.6293 Model for fold 3 saved. Fold 4/5 Epoch 1/10 Train Loss: 1.3514, Train Acc: 0.4748 Val Loss: 1.1433, Val Acc: 0.5642 Epoch 2/10 Train Loss: 1.0219, Train Acc: 0.6131 Val Loss: 1.0011, Val Acc: 0.6268 Epoch 3/10 Train Loss: 0.8253, Train Acc: 0.6934 Val Loss: 0.9803, Val Acc: 0.6440 Epoch 4/10 Train Loss: 0.6547, Train Acc: 0.7631 Val Loss: 1.0478, Val Acc: 0.6350 Epoch 5/10 Train Loss: 0.4808, Train Acc: 0.8293 Val Loss: 1.0787, Val Acc: 0.6537 Epoch 6/10 Train Loss: 0.3519, Train Acc: 0.8770 Val Loss: 1.2807, Val Acc: 0.6478 Epoch 7/10 Train Loss: 0.2559, Train Acc: 0.9121 Val Loss: 1.4376, Val Acc: 0.6445 Epoch 8/10 Train Loss: 0.1808, Train Acc: 0.9385 Val Loss: 1.4842, Val Acc: 0.6609 Epoch 9/10 Train Loss: 0.1380, Train Acc: 0.9526 Val Loss: 1.6784, Val Acc: 0.6580 Epoch 10/10 Train Loss: 0.1162, Train Acc: 0.9606 Val Loss: 1.7237, Val Acc: 0.6503 Model for fold 4 saved. Fold 5/5 Epoch 1/10 Train Loss: 1.3533, Train Acc: 0.4768 Val Loss: 1.1481, Val Acc: 0.5746 Epoch 2/10 Train Loss: 1.0233, Train Acc: 0.6126 Val Loss: 1.0072, Val Acc: 0.6179 Epoch 3/10 Train Loss: 0.8360, Train Acc: 0.6900 Val Loss: 1.0046, Val Acc: 0.6327 Epoch 4/10 Train Loss: 0.6576, Train Acc: 0.7619 Val Loss: 1.0122, Val Acc: 0.6540 Epoch 5/10 Train Loss: 0.4901, Train Acc: 0.8273 Val Loss: 1.0204, Val Acc: 0.6718 Epoch 6/10 Train Loss: 0.3239, Train Acc: 0.8886 Val Loss: 1.2694, Val Acc: 0.6509 Epoch 7/10 Train Loss: 0.2390, Train Acc: 0.9178 Val Loss: 1.2578, Val Acc: 0.6598 Epoch 8/10 Train Loss: 0.1758, Train Acc: 0.9419 Val Loss: 1.5014, Val Acc: 0.6458 Epoch 9/10 Train Loss: 0.1432, Train Acc: 0.9520 Val Loss: 1.5382, Val Acc: 0.6648 Epoch 10/10 Train Loss: 0.0999, Train Acc: 0.9667 Val Loss: 1.6413, Val Acc: 0.6754 Model for fold 5 saved.
In [122]:
vgg_reloaded.save('vgg_reloaded_30epochs.keras')
print('Model saved successfully')
Model saved successfully
In [124]:
from keras.utils import plot_model
plot_model(vgg_reloaded, to_file='model_plot.png', show_shapes=True, show_layer_names=True)
Out[124]:
In [125]:
import matplotlib.pyplot as plt
# Retrieve metrics from the history object
train_loss = history.history['loss']
val_loss = history.history['val_loss']
train_accuracy = history.history['accuracy']
val_accuracy = history.history['val_accuracy']
epochs = range(1, len(train_loss) + 1)
# Plot Loss
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(epochs, train_loss, label='Training Loss', color='blue')
plt.plot(epochs, val_loss, label='Validation Loss', color='orange')
plt.title('Training and Validation Loss for VGG16')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.grid(True)
# Plot Accuracy
plt.subplot(1, 2, 2)
plt.plot(epochs, train_accuracy, label='Training Accuracy', color='blue')
plt.plot(epochs, val_accuracy, label='Validation Accuracy', color='orange')
plt.title('Training and Validation Accuracy for VGG16')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.grid(True)
# Show the plots
plt.tight_layout()
plt.show()
In [119]:
mob_reloaded = tf.keras.models.load_model('50epochs_mobilenet.keras')
print('Model loaded successfully')
Model loaded successfully
In [127]:
# Compile the model
mob_reloaded.compile(
optimizer=Adam(learning_rate=0.0001),
loss='categorical_crossentropy',
metrics=['accuracy']
)
# Define callbacks
callbacks = [
EarlyStopping(
monitor='val_loss',
patience=5,
restore_best_weights=True
),
ReduceLROnPlateau(
monitor='val_loss',
factor=0.2,
patience=3,
verbose=1
)
]
# Training the Model
history = mob_reloaded.fit(
train_generator2,
epochs=20,
validation_data=validation_generator1,
callbacks = callbacks
)
# Evaluate the model
loss, accuracy = mob_reloaded.evaluate(validation_generator1)
print(f"Validation Accuracy: {accuracy * 100:.2f}%")
# Save the trained model
mob_reloaded.save('20_epochs_mobilenet_reloaded.keras')
print('Model saved as 20_epochs_model.keras')
Epoch 1/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 533s 1s/step - accuracy: 0.8241 - loss: 0.4850 - val_accuracy: 0.7135 - val_loss: 0.8259 - learning_rate: 1.0000e-04 Epoch 2/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 529s 1s/step - accuracy: 0.8255 - loss: 0.4789 - val_accuracy: 0.7249 - val_loss: 0.7452 - learning_rate: 1.0000e-04 Epoch 3/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 510s 1s/step - accuracy: 0.8249 - loss: 0.4751 - val_accuracy: 0.7375 - val_loss: 0.7257 - learning_rate: 1.0000e-04 Epoch 4/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 507s 1s/step - accuracy: 0.8286 - loss: 0.4669 - val_accuracy: 0.6951 - val_loss: 0.8551 - learning_rate: 1.0000e-04 Epoch 5/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 507s 1s/step - accuracy: 0.8287 - loss: 0.4711 - val_accuracy: 0.7132 - val_loss: 0.7787 - learning_rate: 1.0000e-04 Epoch 6/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 0s 1s/step - accuracy: 0.8336 - loss: 0.4545 Epoch 6: ReduceLROnPlateau reducing learning rate to 1.9999999494757503e-05. 408/408 ━━━━━━━━━━━━━━━━━━━━ 507s 1s/step - accuracy: 0.8336 - loss: 0.4545 - val_accuracy: 0.6900 - val_loss: 0.8435 - learning_rate: 1.0000e-04 Epoch 7/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 506s 1s/step - accuracy: 0.8501 - loss: 0.4062 - val_accuracy: 0.7425 - val_loss: 0.7111 - learning_rate: 2.0000e-05 Epoch 8/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 506s 1s/step - accuracy: 0.8605 - loss: 0.3778 - val_accuracy: 0.7323 - val_loss: 0.7206 - learning_rate: 2.0000e-05 Epoch 9/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 515s 1s/step - accuracy: 0.8672 - loss: 0.3566 - val_accuracy: 0.7356 - val_loss: 0.7180 - learning_rate: 2.0000e-05 Epoch 10/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 522s 1s/step - accuracy: 0.8685 - loss: 0.3533 - val_accuracy: 0.7571 - val_loss: 0.7091 - learning_rate: 2.0000e-05 Epoch 11/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 515s 1s/step - accuracy: 0.8759 - loss: 0.3471 - val_accuracy: 0.7194 - val_loss: 0.7803 - learning_rate: 2.0000e-05 Epoch 12/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 516s 1s/step - accuracy: 0.8726 - loss: 0.3511 - val_accuracy: 0.7453 - val_loss: 0.7212 - learning_rate: 2.0000e-05 Epoch 13/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 0s 1s/step - accuracy: 0.8772 - loss: 0.3333 Epoch 13: ReduceLROnPlateau reducing learning rate to 3.999999898951501e-06. 408/408 ━━━━━━━━━━━━━━━━━━━━ 517s 1s/step - accuracy: 0.8772 - loss: 0.3333 - val_accuracy: 0.7231 - val_loss: 0.7845 - learning_rate: 2.0000e-05 Epoch 14/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 527s 1s/step - accuracy: 0.8862 - loss: 0.3161 - val_accuracy: 0.7420 - val_loss: 0.7449 - learning_rate: 4.0000e-06 Epoch 15/20 408/408 ━━━━━━━━━━━━━━━━━━━━ 508s 1s/step - accuracy: 0.8859 - loss: 0.3119 - val_accuracy: 0.7428 - val_loss: 0.7486 - learning_rate: 4.0000e-06 110/110 ━━━━━━━━━━━━━━━━━━━━ 30s 275ms/step - accuracy: 0.7624 - loss: 0.7055 Validation Accuracy: 75.71% Model saved as 20_epochs_model.keras
In [128]:
import matplotlib.pyplot as plt
# Retrieve metrics from the history object
train_loss = history.history['loss']
val_loss = history.history['val_loss']
train_accuracy = history.history['accuracy']
val_accuracy = history.history['val_accuracy']
epochs = range(1, len(train_loss) + 1)
# Plot Loss
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(epochs, train_loss, label='Training Loss', color='blue')
plt.plot(epochs, val_loss, label='Validation Loss', color='orange')
plt.title('Training and Validation Loss for MobileNet')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.grid(True)
# Plot Accuracy
plt.subplot(1, 2, 2)
plt.plot(epochs, train_accuracy, label='Training Accuracy', color='blue')
plt.plot(epochs, val_accuracy, label='Validation Accuracy', color='orange')
plt.title('Training and Validation Accuracy for MobileNet')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.grid(True)
# Show the plots
plt.tight_layout()
plt.show()
