Awesome

HRFormer: High-Resolution Transformer for Dense Prediction, NeurIPS 2021

Introduction

This is the official implementation of High-Resolution Transformer (HRFormer). We present a High-Resolution Transformer (HRFormer) that learns high-resolution representations for dense prediction tasks, in contrast to the original Vision Transformer that produces low-resolution representations and has high memory and computational cost. We take advantage of the multi-resolution parallel design introduced in high-resolution convolutional networks (HRNet), along with local-window self-attention that performs self-attention over small non-overlapping image windows, for improving the memory and computation efficiency. In addition, we introduce a convolution into the FFN to exchange information across the disconnected image windows. We demonstrate the effectiveness of the High-Resolution Transformer on human pose estimation and semantic segmentation tasks.

The HRFormer architecture:

teaser

The HRFormer Unit (trans. unit):

teaser

Pose estimation

2d Human Pose Estimation

Results on COCO `val2017` with detector having human AP of 56.4 on COCO `val2017` dataset

Backbone	Input Size	AP	AP<sup>50</sup>	AP<sup>75</sup>	AR<sup>M</sup>	AR<sup>L</sup>	AR	ckpt	log	script
HRFormer-S	256x192	74.0%	90.2%	81.2%	70.4%	80.7%	79.4%	ckpt	log	script
HRFormer-S	384x288	75.6%	90.3%	82.2%	71.6%	82.5%	80.7%	ckpt	log	script
HRFormer-B	256x192	75.6%	90.8%	82.8%	71.7%	82.6%	80.8%	ckpt	log	script
HRFormer-B	384x288	77.2%	91.0%	83.6%	73.2%	84.2%	82.0%	ckpt	log	script

Results on COCO `test-dev` with detector having human AP of 56.4 on COCO `val2017` dataset

Backbone	Input Size	AP	AP<sup>50</sup>	AP<sup>75</sup>	AR<sup>M</sup>	AR<sup>L</sup>	AR	ckpt	log	script
HRFormer-S	384x288	74.5%	92.3%	82.1%	70.7%	80.6%	79.8%	ckpt	log	script
HRFormer-B	384x288	76.2%	92.7%	83.8%	72.5%	82.3%	81.2%	ckpt	log	script

The models are first pre-trained on ImageNet-1K dataset, and then fine-tuned on COCO val2017 dataset.

Semantic segmentation

Cityscapes

Performance on the Cityscapes dataset. The models are trained and tested with input size of 512x1024 and 1024x2048 respectively.

Methods	Backbone	Window Size	Train Set	Test Set	Iterations	Batch Size	OHEM	mIoU	mIoU (Multi-Scale)	Log	ckpt	script
OCRNet	HRFormer-S	7x7	Train	Val	80000	8	Yes	80.0	81.0	log	ckpt	script
OCRNet	HRFormer-B	7x7	Train	Val	80000	8	Yes	81.4	82.0	log	ckpt	script
OCRNet	HRFormer-B	15x15	Train	Val	80000	8	Yes	81.9	82.6	log	ckpt	script

PASCAL-Context

The models are trained with the input size of 520x520, and tested with original size.

Methods	Backbone	Window Size	Train Set	Test Set	Iterations	Batch Size	OHEM	mIoU	mIoU (Multi-Scale)	Log	ckpt	script
OCRNet	HRFormer-S	7x7	Train	Val	60000	16	Yes	53.8	54.6	log	ckpt	script
OCRNet	HRFormer-B	7x7	Train	Val	60000	16	Yes	56.3	57.1	log	ckpt	script
OCRNet	HRFormer-B	15x15	Train	Val	60000	16	Yes	57.6	58.5	log	ckpt	script

COCO-Stuff

The models are trained with input size of 520x520, and tested with original size.

Methods	Backbone	Window Size	Train Set	Test Set	Iterations	Batch Size	OHEM	mIoU	mIoU (Multi-Scale)	Log	ckpt	script
OCRNet	HRFormer-S	7x7	Train	Val	60000	16	Yes	37.9	38.9	log	ckpt	script
OCRNet	HRFormer-B	7x7	Train	Val	60000	16	Yes	41.6	42.5	log	ckpt	script
OCRNet	HRFormer-B	15x15	Train	Val	60000	16	Yes	42.4	43.3	log	ckpt	script

ADE20K

The models are trained with input size of 520x520, and tested with original size. The results with window size 15x15 will be updated latter.

Methods	Backbone	Window Size	Train Set	Test Set	Iterations	Batch Size	OHEM	mIoU	mIoU (Multi-Scale)	Log	ckpt	script
OCRNet	HRFormer-S	7x7	Train	Val	150000	8	Yes	44.0	45.1	log	ckpt	script
OCRNet	HRFormer-B	7x7	Train	Val	150000	8	Yes	46.3	47.6	log	ckpt	script
OCRNet	HRFormer-B	13x13	Train	Val	150000	8	Yes	48.7	50.0	log	ckpt	script
OCRNet	HRFormer-B	15x15	Train	Val	150000	8	Yes	-	-	-	-	-

Classification

Results on ImageNet-1K

Backbone	acc@1	acc@5	#params	FLOPs	ckpt	log	script
HRFormer-T	78.6%	94.2%	8.0M	1.83G	ckpt	log	script
HRFormer-S	81.2%	95.6%	13.5M	3.56G	ckpt	log	script
HRFormer-B	82.8%	96.3%	50.3M	13.71G	ckpt	log	script

Citation

If you find this project useful in your research, please consider cite:

@article{YuanFHLZCW21,
  title={HRFormer: High-Resolution Transformer for Dense Prediction},
  author={Yuhui Yuan and Rao Fu and Lang Huang and Weihong Lin and Chao Zhang and Xilin Chen and Jingdong Wang},
  booktitle={NeurIPS},
  year={2021}
}

Acknowledgment

This project is developed based on the Swin-Transformer, openseg.pytorch, and mmpose.

git diff-index HEAD
git subtree add -P pose <url to sub-repo> <sub-repo branch>