CIFAR-10 CNN Progression

CIFAR-10 contains small color images from ten classes and is more varied than MNIST. This example moves from a simple two-stage convolutional network to the deeper model represented in the source material, adding convolutional stages, normalization, dropout, and a larger classifier.

The same training, validation, and test examples are used throughout so the simple and advanced models can be viewed as one clear progression.

• Complete PyTorch workflow

• Keras 3 alternative using the Torch backend