How to Use TensorFlow for Image Classification: A Step-by-Step Guide

TensorFlow simplifies building image classification models, even for beginners. With its high-level Keras API, you can go from raw images to a trained classifier in just a few lines of code. This tutorial walks through the essential steps using a standard dataset.

First, install TensorFlow (e.g., pip install tensorflow) and load a dataset like CIFAR‑10 or your own folder of images. Use tf.keras.utils.image_dataset_from_directory to load images from subfolders, splitting into training and validation sets. Normalize pixel values to the range [0,1] by scaling with Rescaling(1./255).

Article illustration

1. Build the Model

Create a sequential model with convolutional layers followed by dense layers. A typical structure:

  • Conv2D(32, (3,3), activation='relu') – learns low-level features
  • MaxPooling2D(2,2) – reduces spatial dimensions
  • Repeat the above two layers with 64 filters
  • Flatten() – converts 2D to 1D
  • Dense(128, activation='relu') – fully connected
  • Dense(num_classes, activation='softmax') – outputs probabilities

2. Compile and Train

Compile the model using:

  • optimizer='adam'
  • loss='sparse_categorical_crossentropy' for integer labels
  • metrics=['accuracy']

Train with model.fit(train_ds, validation_data=val_ds, epochs=10). Use callbacks like EarlyStopping to prevent overfitting.

3. Evaluate and Predict

After training, evaluate on the test set: model.evaluate(test_ds). For a new image, load it as a tensor, expand its batch dimension, and call model.predict(img_tensor). The argmax of the output gives the predicted class.

TensorFlow makes image classification accessible and scalable. Start with a simple CNN, then experiment with transfer learning (e.g., MobileNetV2) to boost accuracy on custom datasets.

sarah antaboga
Author: sarah antaboga

Leave a Reply

Your email address will not be published. Required fields are marked *