mmyolo

Why we included this project

MMYOLO is a solid starting point when you need to train, fine-tune, or benchmark a modern real-time object detector. It gathers the popular YOLO-family models, YOLOv5 through YOLOv8, YOLOX, PPYOLOE, and the team's own RTMDet, into one modular PyTorch codebase built on MMDetection. Sharing that foundation means you can swap backbones, heads, and training recipes as interchangeable components and run fair comparisons on the same benchmark, instead of juggling separate repos with different APIs. It also covers rotated object detection, which is handy for aerial and satellite imagery, plus instance segmentation, with pretrained checkpoints and tutorials for newcomers and teams planning to deploy. The unified recipe system is the main draw if you want reproducible, well-documented training without reinventing the pipeline.