[Feature] Add Swin Large(Swin-L) Transformer models

[MengzhangLI]

Motivation

Add Swin-L transformer models, they are not listed in ADE20K results in original paper/repo but are widely used for comparison between current backbone models.

Results and models

Method	Backbone	Crop Size	pretrain	pretrain img size	Batch Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)
UperNet	Swin-T	512x512	ImageNet-1K	224x224	16	160000	5.02	21.06	44.41	45.79
UperNet	Swin-S	512x512	ImageNet-1K	224x224	16	160000	6.17	14.72	47.72	49.24
UperNet	Swin-B	512x512	ImageNet-1K	224x224	16	160000	7.61	12.65	47.99	49.57
UperNet	Swin-B	512x512	ImageNet-22K	224x224	16	160000	-	-	50.31	51.9
UperNet	Swin-B	512x512	ImageNet-1K	384x384	16	160000	8.52	12.10	48.35	49.65
UperNet	Swin-B	512x512	ImageNet-22K	384x384	16	160000	-	-	50.76	52.4
UperNet	Swin-L	512x512	ImageNet-22K	224x224	16	160000	10.98	8.23	51.17	52.99
UperNet	Swin-L	512x512	ImageNet-22K	384x384	16	160000	12.42	7.57	52.25	54.12

[codecov]

Codecov Report

Merging #1471 (0e7cafe) into master (e3b5b5e) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #1471   +/-   ##
=======================================
  Coverage   89.04%   89.04%           
=======================================
  Files         144      144           
  Lines        8636     8636           
  Branches     1458     1458           
=======================================
  Hits         7690     7690           
  Misses        706      706           
  Partials      240      240           

Flag	Coverage Δ
unittests	89.04% <ø> (ø)

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

---

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 e3b5b5e...0e7cafe. Read the comment docs.