twins

Introduction

Official Repo

Code Snippet

Twins (NeurIPS'2021)

@article{chu2021twins,
    title={Twins: Revisiting spatial attention design in vision transformers},
    author={Chu, Xiangxiang and Tian, Zhi and Wang, Yuqing and Zhang, Bo and Ren, Haibing and Wei, Xiaolin and Xia, Huaxia and Shen, Chunhua},
    journal={arXiv preprint arXiv:2104.13840},
    year={2021}
}

Results

ADE20k

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU	Download
UperNet	ImageNet-1k-224x224	SVT-S	512x512	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	46.16%	cfg | model | log
UperNet	ImageNet-1k-224x224	SVT-B	512x512	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	48.05%	cfg | model | log
UperNet	ImageNet-1k-224x224	SVT-L	512x512	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	49.80%	cfg | model | log
UperNet	ImageNet-1k-224x224	PCPVT-S	512x512	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	46.07%	cfg | model | log
UperNet	ImageNet-1k-224x224	PCPVT-B	512x512	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	48.06%	cfg | model | log
UperNet	ImageNet-1k-224x224	PCPVT-L	512x512	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	49.35%	cfg | model | log

More

You can also download the model weights from following sources:

• BaiduNetdisk: https://pan.baidu.com/s/1gD-NJJWOtaHCtB0qHE79rA with access code s757