Awesome-Backbones

Why we included this project

This project collects PyTorch implementations of a broad set of image classification backbones, from classic LeNet and ResNet to newer EfficientNet, ConvNeXt, Swin Transformer, DeiT, and MobileViT. Each model comes with its own config file, so you can train, evaluate, and swap backbones without rewriting the surrounding pipeline. The tutorials walk through environment setup, dataset preparation, config explanation, and adding new model components, which helps students and engineers understand the internals rather than just calling a pretrained API. It also includes scripts for computing FLOPs and parameters, visualizing class activation maps, and exporting to ONNX, making it a handy comparison and experimentation toolkit. For teams working with small custom datasets, the practical notes on tuning learning rate, batch size, and data augmentation are a real time-saver.