ultralyticsPro

Why we included this project

Anyone who trains or fine-tunes YOLO-style detectors will find this repo a handy base for testing architecture changes. It doesn't ship a single fixed model; instead it gathers multiple YOLO versions (v5, v7, v8, v10, v11, RT-DETR) into one PyTorch codebase, letting you swap backbones, necks, heads, attention blocks, loss functions, or NMS logic without touching the rest of the training pipeline. The modular design suits people who want quick experiments and side-by-side comparisons on their own data, whether for a paper or for squeezing out accuracy in a real deployment. It also covers segmentation, classification, and pose estimation, so it's useful beyond plain detection.