MMDetection

OpenMMLab's MMDetection is one of the standard research toolboxes for computer vision detection work, built on PyTorch and covering object detection, instance segmentation, panoptic segmentation, and semi-supervised detection. Its modular design decomposes the detection pipeline into interchangeable components, so backbones, necks, heads, and losses can be recombined into custom architectures through configuration files rather than rewritten code. The model zoo is extensive, with hundreds of pretrained checkpoints across dozens of algorithm families including Faster R-CNN, Cascade R-CNN, RetinaNet, DETR variants, YOLO-style detectors, and the RTMDet series, backed by backbones from ResNet to Vision Transformers and ConvNeXt. It sits on the MMEngine and MMCV libraries shared across the OpenMMLab ecosystem and requires PyTorch 1.8 or newer. Released under Apache 2.0, it is used in academic papers as a reproducible baseline platform and in industry for building production detection systems from vetted implementations.