support transformer backbone

[xiexinch]

Support transformer backbone.
This PR modified from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py.

• Support load pretrained checkpoints from reference repo.
• Training image size can be different from pretrain image.
• Support dynamic test image size at evaluation.

Usage

Backbone config should be like this.

backbone=dict(
        type='VisionTransformer',
        img_size=(512, 512),
        patch_size=16,
        in_channels=3,
        embed_dim=768,
        depth=12,
        mlp_ratio=4,
        qkv_bias=True,
        qk_scale=None,
        representation_size=None,
        drop_rate=0.0,
        attn_drop_rate=0.0,
        norm_cfg=dict(type='LN'),
        act_cfg=dict(type='GELU'),
        norm_eval=False)

We can download checkpoints from https://github.com/rwightman/pytorch-image-models by setting pretrained

model = dict(
pretrained='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_p16_224-80ecf9dd.pth',
backbone=dict(...))

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@67eee62). Click here to learn what that means.
The diff coverage is 88.05%.

@@            Coverage Diff            @@
##             master     #465   +/-   ##
=========================================
  Coverage          ?   86.53%           
=========================================
  Files             ?       98           
  Lines             ?     5133           
  Branches          ?      829           
=========================================
  Hits              ?     4442           
  Misses            ?      534           
  Partials          ?      157           

Flag	Coverage Δ
unittests	86.53% <88.05%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmseg/models/backbones/vit.py	87.97% <87.97%> (ø)
mmseg/models/backbones/__init__.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 67eee62...6827803. Read the comment docs.

[xvjiarui]

There are some unresolved comments.