Reproduce MIM pre-training (v0.2.1)

[Lupin1998]

Reproduce MIM pre-training methods in OpenMixup to v0.2.1:

1. Support MAE and SimMIM according to official implementation and MMSelfsup, including pre-training, benchmarks, visualization, etc. Results of MIM methods will be released soon.
2. Unofficial implementation of MaskFeat according to their paper.
3. Fix bugs in regression metrics, MIM dataset, and benchmark configs.

Currently, we are working on SSL methods, the update of supervised mixup methods (e.g., adding new mixup variants) might stagnate for two weeks. Besides, we suggest you install PyTorch 1.8 or higher (required by some self-supervised methods) and mmcv-full for better usage of this repo. You can still use PyTorch 1.6 for supervised classification methods.

[Lupin1998]

OpenMixup v0.2.1 is available and bugs are fixed.