[Feature] Add PoolFormer (CVPR'2022)

[MengzhangLI]

Motivation

Support PoolFormer (CVPR'2022)
Paper: MetaFormer is Actually What You Need for Vision
Code: https://github.com/sail-sg/poolformer/tree/main/segmentation
Upstream backbone from MMClassification: https://github.com/open-mmlab/mmclassification/tree/master/configs/poolformer

• Align Inference Results against official repo
• Align Training metric of models to repeat results
• Post-processing: Upload README.md and models & logs.

Results about Aligning Inference metric

Method	Backbone	Pretrain	Iters	mIoU
Semantic FPN	PoolFormer-S12	ImageNet-1K	40K	37.15
Semantic FPN	PoolFormer-S24	ImageNet-1K	40K	40.27
Semantic FPN	PoolFormer-S36	ImageNet-1K	40K	41.97
Semantic FPN	PoolFormer-M36	ImageNet-1K	40K	42.36
Semantic FPN	PoolFormer-M48	ImageNet-1K	40K	42.72

Thus, the inference metric has been aligned, which means the model structure and test pipeline are both totally correct.

Results about Aligning Training metric

Method	Backbone	Pretrain	Iters	mIoU (paper)	re-training
Semantic FPN	PoolFormer-S12	ImageNet-1K	40K	37.15	37.07
Semantic FPN	PoolFormer-S24	ImageNet-1K	40K	40.27	40.51
Semantic FPN	PoolFormer-S36	ImageNet-1K	40K	41.97	41.81
Semantic FPN	PoolFormer-M36	ImageNet-1K	40K	42.36	42.35
Semantic FPN	PoolFormer-M48	ImageNet-1K	40K	42.72	42.43

Thus, PoolFormer could be re-implemented in this PR.

Total Results

Method	Backbone	Crop Size	pretrain	Batch Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	mIoU*	mIoU*(ms+flip)	config	download
FPN	PoolFormer-S12	512x512	ImageNet-1K	32	40000	4.17	23.48	36.0	36.42	37.07	38.44	config	model | log
FPN	PoolFormer-S24	512x512	ImageNet-1K	32	40000	5.47	15.74	39.35	39.73	40.36	41.08	config	model | log
FPN	PoolFormer-S36	512x512	ImageNet-1K	32	40000	6.77	11.34	40.64	40.99	41.81	42.72	config	model | log
FPN	PoolFormer-M36	512x512	ImageNet-1K	32	40000	8.59	8.97	40.91	41.28	42.35	43.34	config	model | log
FPN	PoolFormer-M48	512x512	ImageNet-1K	32	40000	10.48	6.69	41.82	42.2	42.76	43.57	config	model | log

Note:

• mIoU with * is collected when ResizeToMultiple is adopted in test_pipeline, so do mIoU in logs.

[codecov]

Codecov Report

Base: 89.11% // Head: 89.14% // Increases project coverage by +0.03% 🎉

Coverage data is based on head (bbdac5c) compared to base (ee25adc).
Patch has no changes to coverable lines.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1537      +/-   ##
==========================================
+ Coverage   89.11%   89.14%   +0.03%     
==========================================
  Files         145      145              
  Lines        8716     8716              
  Branches     1471     1471              
==========================================
+ Hits         7767     7770       +3     
+ Misses        707      704       -3     
  Partials      242      242              

Flag	Coverage Δ
unittests	89.14% <ø> (+0.03%)	⬆️

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

Impacted Files	Coverage Δ
mmseg/datasets/pipelines/transforms.py	98.64% <0.00%> (+0.58%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.