EasyCV

Why we included this project

For teams juggling several vision workloads, EasyCV keeps everything in one PyTorch codebase. It ships training pipelines and pretrained weights for self-supervised contrastive learning (SimCLR, MoCo V2, SwAV, DINO) and masked image modeling (MAE), so you can pretrain or fine-tune without stitching together separate repositories. It also covers ordinary tasks such as image classification, object detection, and metric learning, leaning on transformer backbones like ViT, Swin, and the DETR family. The framework splits into modular components (dataset, model, training hooks), which makes it easy to add a new dataset or swap a model for an experiment. With a command-line and config-based interface and support for many timm checkpoints, it suits researchers prototyping supervised and self-supervised models as well as engineers running standard CV benchmarks.