pytorch-cifar100

Why we included this project

This is a practical PyTorch codebase for CIFAR-100 that packs in clean implementations of a wide range of convolutional architectures: ResNet, DenseNet, VGG, the Inception variants, ResNeXt, MobileNet, SqueezeNet, SENet, NASNet, WideResNet, and more. Because every model trains under the same hyperparameters and reports accuracy the same way, it is easy to compare how different CNN designs actually behave on one benchmark. Switching models means changing a single flag on the train.py command line, and the per-model error tables give a realistic baseline before you commit GPU time. For anyone who wants a small, self-contained starting point for an image classification experiment instead of a full framework, this also reads well as an example of how each architecture goes from paper to working code.