Iteration Plan - Mar 2021

[dreamerlin]

Since MMAction2 v0.12.0 has been released, now it's time to plan for the next iteration, i.e. v0.13.0.
This will be a one-month iteration and our focus in this iteration is to support more algorithms and refactor some codes.

Algorithms:

• LFB: Lone-Term Feature Bank ([Feature] Support LFB #553)
  Video Transformer researching

Tools:

• MMCls/TorchVision Backbone registery ([Feature] Support MMCls backbones for TSN  #679, [Feature] Enable torchvision backbones #720)
  Support auto_fp16 using torch.cuda.amp

Misc:

• CN Tutorial ([Docs] CN tutorials #665)
• Support CN Readthedocs construction ([Docs] Build Chinese Docs #716)
• Multi-repo registry usage ([Docs] CN Demo #717, [Feature] Enable torchvision backbones #720)
  Entrypoint Refactoring
  Unify parameters initilization
  Multi-repo registry usage ([Docs] CN Demo #717, [Feature] Enable torchvision backbones #720)

---

Deferred

• Entrypoint Refactoring
• Unify parameters initilization
• Adapt Registry conflict resolution
• Video Transformer researching
• Support auto_fp16 using torch.cuda.amp

[irvingzhang0512]

• mixup/cutmix [Feature] Support Mixup and Cutmix for Recognizers. #681
• recognizers(pytorch/onnxruntime/tensorrt) latency Comparisons for Recognizers #773
• support onnx/tensorrt model for test.py [Improvement] Test with onnx models and TensorRT engines. #758

[Bo396543018]

Hope to add these methods which i found useful for some competitions

• TEA: Temporal Excitation and Aggregation for Action Recognition (CVPR2020) https://github.com/Phoenix1327/tea-action-recognition
• PAN: Persistent Appearance Network: https://github.com/zhang-can/PAN-PyTorch
• TDN: Temporal Difference Networks for Efficient Action Recognition (CVPR 2021) https://github.com/MCG-NJU/TDN

[ahkarami]

Also, hope to add at least one Temporal Activity Segmentation method (such as BCN, SSTDA, etc.)