- Vision Transformer backbone, see config in
configs/Market1501/bagtricks_vit.yml
- Self-Distillation with EMA update
- Gradient Clip
- Faster dataloader with pre-fetch thread and cuda stream
- Optimize DDP training speed by removing
find_unused_parameters
in DDP
- Multiple machine training support
- RepVGG backbone
- Partial FC
- Torch2trt pipeline
- Decouple linear transforms and softmax
- config decorator
- NAIC20(reid track) 1-st solution
- Multi-teacher Knowledge Distillation
- TRT network definition APIs in FastRT