lucasjinreal/yolov7_d2

Integrates multiple detection architectures (DETR, YOLOX, anchor-free variants) and vision transformers (Swin, PVT, ConvNext) within Detectron2's framework, enabling seamless backbone swapping and multi-task extensions like keypoint detection and instance segmentation. Supports ONNX export and TensorRT deployment for transformer-based models, with built-in training utilities including advanced augmentations (Mosaic, GridMask) and loss functions (CIoU, GFL) from recent SOTA detectors.

3,114 stars. No commits in the last 6 months.