CNN Image Classification Pipeline

Create a complete CNN-based image classification system.

Project details:
- Number of classes: [N_CLASSES]
- Image dimensions: [HEIGHT x WIDTH x CHANNELS]
- Dataset size: [APPROXIMATE NUMBER OF IMAGES]
- Framework preference: [PyTorch/TensorFlow/Keras]

Deliverables:
1. Data loading with augmentation (rotation, flip, crop, color jitter)
2. CNN architecture with:
   - Convolutional layers with batch normalization
   - Pooling strategies
   - Dropout regularization
   - Global average pooling
3. Transfer learning option with pretrained backbones
4. Training with mixed precision
5. Learning rate finder
6. Confusion matrix and classification report
7. Grad-CAM visualization for interpretability
8. Model export for deployment

Handle class imbalance with weighted loss or oversampling.