End to End Projects - Flower Classification Project Deep

ResNet

import tensorflow as tf
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.applications import ResNet50, MobileNetV2
from tensorflow.keras.applications.resnet50 import preprocess_input as resnet_preprocess
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input as mobilenet_preprocess
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Flatten, GlobalAveragePooling2D
import matplotlib.pyplot as plt
import tensorflow_datasets as tfds

# Load Oxford Flowers 102 dataset
(dataset_train, dataset_test), dataset_info = tfds.load(
    'oxford_flowers102',
    split=['train', 'test'],
    as_supervised=True,
    with_info=True
)

# Dataset preprocessing
def preprocess(image, label, model_type='resnet'):
    image = tf.image.resize(image, (224, 224))  # Resize for input to models
    if model_type == 'resnet':
        image = resnet_preprocess(image)  # Preprocess for ResNet
    elif model_type == 'mobilenet':
        image = mobilenet_preprocess(image)  # Preprocess for MobileNet
    return image, label

# Apply preprocessing
batch_size = 32
train_dataset = dataset_train.map(lambda x, y: preprocess(x, y, model_type='resnet')).batch(batch_size).prefetch(tf.data.AUTOTUNE)
test_dataset = dataset_test.map(lambda x, y: preprocess(x, y, model_type='resnet')).batch(batch_size).prefetch(tf.data.AUTOTUNE)

# Display a few samples
fig, axes = plt.subplots(1, 5, figsize=(15, 5))
for i, (image, label) in enumerate(dataset_train.take(5)):
    axes[i].imshow(image)
    axes[i].set_title(f"Label: {label}")
    axes[i].axis('off')
plt.tight_layout()
plt.show()

Downloading and preparing dataset 328.90 MiB (download: 328.90 MiB, generated: 331.34 MiB, total: 660.25 MiB) to /root/tensorflow_datasets/oxford_flowers102/2.1.1...
Dataset oxford_flowers102 downloaded and prepared to /root/tensorflow_datasets/oxford_flowers102/2.1.1. Subsequent calls will reuse this data.

from tensorflow.keras.optimizers import Adam

# Load ResNet50 pre-trained on ImageNet
resnet_base = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# Freeze base layers
resnet_base.trainable = False

# Add custom layers
resnet_model = Sequential([
    resnet_base,
    GlobalAveragePooling2D(),  # Reduces feature maps to a single vector
    Dense(256, activation='relu'),
    Dropout(0.5),
    Dense(102, activation='softmax')  # 102 flower categories
])

# Compile the model
resnet_model.compile(optimizer=Adam(learning_rate=0.001), loss='sparse_categorical_crossentropy', metrics=['accuracy'])

# Train the model
resnet_history = resnet_model.fit(train_dataset, epochs=10, validation_data=test_dataset)

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
94765736/94765736 ━━━━━━━━━━━━━━━━━━━━ 1s 0us/step
Epoch 1/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1223s 39s/step - accuracy: 0.0264 - loss: 4.9411 - val_accuracy: 0.1935 - val_loss: 4.0526
Epoch 2/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1206s 39s/step - accuracy: 0.1903 - loss: 3.6557 - val_accuracy: 0.4110 - val_loss: 2.9810
Epoch 3/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1204s 39s/step - accuracy: 0.4168 - loss: 2.5418 - val_accuracy: 0.5807 - val_loss: 2.1371
Epoch 4/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1275s 41s/step - accuracy: 0.5482 - loss: 1.8559 - val_accuracy: 0.6678 - val_loss: 1.7028
Epoch 5/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1215s 39s/step - accuracy: 0.6594 - loss: 1.3788 - val_accuracy: 0.7113 - val_loss: 1.3727
Epoch 6/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1229s 39s/step - accuracy: 0.7320 - loss: 1.0761 - val_accuracy: 0.7447 - val_loss: 1.2135
Epoch 7/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1328s 41s/step - accuracy: 0.7958 - loss: 0.8285 - val_accuracy: 0.7671 - val_loss: 1.0742
Epoch 8/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1218s 39s/step - accuracy: 0.8369 - loss: 0.6474 - val_accuracy: 0.7837 - val_loss: 0.9577
Epoch 9/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1274s 41s/step - accuracy: 0.8738 - loss: 0.4881 - val_accuracy: 0.7935 - val_loss: 0.9017
Epoch 10/10
32/32 ━━━━━━━━━━━━━━━━━━━━ 1213s 39s/step - accuracy: 0.9052 - loss: 0.3989 - val_accuracy: 0.7923 - val_loss: 0.8866

# Evaluate the model
loss, accuracy = resnet_model.evaluate(test_dataset)
print(f"Test Accuracy: {accuracy * 100:.2f}%")

193/193 ━━━━━━━━━━━━━━━━━━━━ 1064s 6s/step - accuracy: 0.7924 - loss: 0.8885
Test Accuracy: 79.23%

import numpy as np
import matplotlib.pyplot as plt

# Get class names from the dataset info
class_names = dataset_info.features['label'].names

# Select a batch of images and labels from the test dataset
test_batch = next(iter(test_dataset))
test_images, true_labels = test_batch

# Get predictions for the batch
predictions = resnet_model.predict(test_images)
predicted_labels = np.argmax(predictions, axis=1)

# Display the first 5 images in the batch
fig, axes = plt.subplots(1, 5, figsize=(20, 5))
for i, ax in enumerate(axes):
    # Display the image
    ax.imshow(test_images[i].numpy().astype('uint8'))

    # Set the title with true and predicted labels
    true_label = class_names[true_labels[i]]
    predicted_label = class_names[predicted_labels[i]]
    ax.set_title(f"True: {true_label}\nPredicted: {predicted_label}", fontsize=12)

    ax.axis('off')  # Hide axes for better visualization

plt.tight_layout()
plt.show()

1/1 ━━━━━━━━━━━━━━━━━━━━ 7s 7s/step