How can Tensorflow be used to extract features with the help of pre-trained model using Python?

TensorFlow can be used to extract features with the help of pre-trained models using a feature extractor model, which is previously defined and is used in the KerasLayer method. This approach leverages transfer learning to utilize knowledge from models trained on large datasets.

Read More: What is TensorFlow and how Keras work with TensorFlow to create Neural Networks?

Understanding Transfer Learning

The intuition behind transfer learning for image classification is that if a model is trained on a large and general dataset, this model can effectively serve as a generic model for the visual world. It learns feature maps, which means users don't have to start from scratch by training a large model on a large dataset.

TensorFlow Hub is a repository that contains pre-trained TensorFlow models. TensorFlow can be used to fine-tune learning models.

Setting Up the Environment

First, we need to import the required libraries and set up our environment:

import tensorflow as tf
import tensorflow_hub as hub
import numpy as np

# Create sample image batch (32 images of 224x224x3)
image_batch = tf.random.normal((32, 224, 224, 3))
print("Image batch shape:", image_batch.shape)

Image batch shape: (32, 224, 224, 3)

Extracting Features Using Pre-trained Model

We use MobileNet V2 from TensorFlow Hub as our feature extractor. This pre-trained model converts images into feature vectors:

print("Extracting features")
feature_extractor_model = "https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/4"

# Create the feature extractor layer
feature_extractor_layer = hub.KerasLayer(
    feature_extractor_model, 
    input_shape=(224, 224, 3), 
    trainable=False
)

# Extract features from the image batch
feature_batch = feature_extractor_layer(image_batch)
print("Feature batch shape:", feature_batch.shape)

Extracting features
Feature batch shape: (32, 1280)

How Feature Extraction Works

The feature extractor transforms each input image (224×224×3) into a 1280-dimensional feature vector. These features represent high-level patterns learned by the pre-trained model:

• Input: 32 images of shape (224, 224, 3)
• Output: 32 feature vectors of shape (1280,)
• The model is frozen (trainable=False) to preserve pre-trained weights

Using Features for Classification

These extracted features can be used as input to a custom classifier:

# Create a simple classification model using extracted features
model = tf.keras.Sequential([
    feature_extractor_layer,
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')  # 10 classes
])

# Compile the model
model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)

print("Model summary:")
model.summary()

Model summary:
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
keras_layer (KerasLayer)     (None, 1280)              2257984   
_________________________________________________________________
dense (Dense)                (None, 128)               163968    
_________________________________________________________________
dense_1 (Dense)              (None, 10)                1290      
=================================================================
Total params: 2,423,242
Trainable params: 165,258
Non-trainable params: 2,257,984
_________________________________________________________________

Conclusion

TensorFlow Hub's pre-trained models enable efficient feature extraction without training from scratch. The KerasLayer wrapper makes it easy to integrate pre-trained models into custom architectures, significantly reducing training time and computational requirements.