[Fix] Support MaskFormer.

[wanghao9610]

Motivation

Add MaskFormer to mmseg.

Modification

• add 'ToMask' formatting operator to train pipeline to generate the 'gt_masks' and 'gt_labels'.
• change 'DefaultFormatBundle' class to adapt to new keys ' gt_masks' and 'gt_labels'.
• change 'EncoderDecoder' forward_train input elements (adding **kwargs to match the maskformer input demand).
• add 'maskformer_head', reference with the implementation of mmdet.
• change 'dice_loss' and 'focal_loss' reference with them on mmdet, the original dice_loss will lead the out_of_memory error because of the F.one_hot function when num_class is large.
• add 'plugin' subdirectory simply adding other modules, such as PixelDecoder, TransformerEncodePixelDecoder.
• add 'MaskHungarianAssigner' to do mask bipartite matching, reference with mmdet.
• add 'positional_encoding' and 'transformer', reference with mmdet.
• add some configs of maskformer:
  • maskformer_r50_512x512_80k_ade20k.py
  • maskformer_r50_512x512_160k_ade20k.py
  • maskformer_swin-t-p4-w7_512x512_160k_ade20k.py

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

wanghao seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[MengzhangLI]

Based on our inner discussion, we would like to use MaskFormer module directly from MMDetection, rather than making a lot of code copy to MMSegmentation. We would support MaskFormer as soon as possible.

[wmkai]

Hi, is there any plan to support Mask2former in semantic segmentation?

[OpenMMLab-Assistant003]

Hi @wanghao9610！We are grateful for your efforts in helping improve this open-source project during your personal time. Welcome to join OpenMMLab Special Interest Group (SIG) private channel on Discord, where you can share your experiences, ideas, and build connections with like-minded peers. To join the SIG channel, simply message moderator— OpenMMLab on Discord or briefly share your open-source contributions in the #introductions channel and we will assist you. Look forward to seeing you there! Join us ：https://discord.gg/UjgXkPWNqA If you have a WeChat account，welcome to join our community on WeChat. You can add our assistant ：openmmlabwx. Please add "mmsig + Github ID" as a remark when adding friends：） Thank you again for your contribution❤ @wanghao9610