Detectron2 is Facebook AI Research's next generation software system that implements state-of-the-art object detection algorithms. Official address
- detailed example and data help you to play every models in detectron2.
- add more data augmentation, visualization in training.
- add more models.
- add recent trick.
- deploy your model (ORT, TensorRT, libtorch).
See INSTALL.md
build from source.
See EASY_START.md
- CBAM
- ...
- GradCam++
- ...
- voc split split voc data to train,val and test set
- generate file list and label generate two texts with image list and label list respectively
- voc to coco modify from voc2coco convert voc to coco format
- analyze data output the basic information of training data
- json_to_txt convert coco to txt for some yolo models
- json_to_png convert coco to png for semantic models
- coco_split.py split coco data to train,val and test set
- ...
- FCOS
- CenterMask2
- DETR
- YOLO
- ...
- FasterRCNN(C4)-tensorrt
- MaskRCNN(C4)-tensorrt
- RotatedRCNN(C4)-tensorrt
- RetinaNet-tensorrt refer to NVIDIA
- FCOS-tensorrt
- CenterMask2-tensorrt
- DETR-tensorrt
- ...
- ABCPruner
- ...