Iteration Plan - Feb 2021

[dreamerlin]

Happy Chinese New Year!

Since MMAction2 v0.11.0 has been released, now it's time to plan for the next iteration, i.e. v0.12.0.
This will be a one-month iteration and our focus in this iteration is to support more algorithm and refactor some codes.

Algorithms:

• TAM ([Feature] Support TAM #595)
• Support TSM-MobileNetV2 ([Feature] Support TSM-MobileNetV2 #415)

Improvement

• Support Imgaug for augmentations in the data pipeline ([Feature] Support Imgaug for augmentations in the data pipeline. #492)
• Train/Test custom classes for ava dataset ([Improvement] Training custom classes of ava dataset #555)
• Support soft label for CrossEntropyLoss ([Improvement] Support soft label for CrossEntropyLoss #625)

Misc

• GPU Normalize ([Feature] Support GPU Normalize #586)
• Support flip with label mapping ([Feature] Support flip with label mapping #591)
• CN README (Update README_cn.md #594)
• Config Refactor ([Refactor] Config Refactor (train_cfg & test_cfg) #629)

delay:
Long-term feature banks
Add model registry
Support auto_fp16 using torch.cuda.amp