TASK	DATASET	MODEL	METRIC NAME	METRIC VALUE	GLOBAL RANK
Semantic Segmentation	ADE20K	ResNeSt-269	Validation mIoU	47.60	# 54
Semantic Segmentation	ADE20K	ResNeSt-101	Validation mIoU	46.91	# 58
Semantic Segmentation	ADE20K	ResNeSt-200	Validation mIoU	48.36	# 45
Semantic Segmentation	ADE20K val	ResNeSt-101	mIoU	46.91	# 45
Semantic Segmentation	ADE20K val	ResNeSt-200	mIoU	48.36	# 40
Semantic Segmentation	ADE20K val	ResNeSt-269	mIoU	47.60	# 42
Semantic Segmentation	Cityscapes test	ResNeSt200	Mean IoU (class)	83.3	# 9
Semantic Segmentation	Cityscapes val	ResNeSt-200	mIoU	82.7	# 11
Object Detection	COCO minival	ResNeSt-200 (single-scale)	box AP	50.54	# 38
Object Detection	COCO minival	ResNeSt-200 (single-scale)	AP50	68.78	# 17
Object Detection	COCO minival	ResNeSt-200 (single-scale)	AP75	55.17	# 12
Object Detection	COCO minival	ResNeSt-200 (single-scale)	APM	54.2	# 13
Object Detection	COCO minival	ResNeSt-200 (single-scale)	APL	63.9	# 13
Instance Segmentation	COCO minival	ResNeSt-200 (multi-scale)	mask AP	46.25	# 19
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	box AP	50.91	# 36
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	AP50	69.53	# 14
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	AP75	55.40	# 10
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	APS	32.67	# 12
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	APM	54.66	# 12
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	APL	65.83	# 11
Instance Segmentation	COCO minival	ResNeSt-200-DCN (single-scale)	mask AP	44.5	# 27
Panoptic Segmentation	COCO minival	PanopticFPN+ResNeSt(single-scale)	PQ	47.9	# 6
Panoptic Segmentation	COCO minival	PanopticFPN+ResNeSt(single-scale)	PQth	55.1	# 7
Panoptic Segmentation	COCO minival	PanopticFPN+ResNeSt(single-scale)	PQst	37.0	# 6
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	box AP	52.47	# 31
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	AP50	71.00	# 8
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	AP75	57.07	# 7
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	APS	36.80	# 8
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	APM	56.36	# 9
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	APL	66.29	# 10
Instance Segmentation	COCO minival	ResNeSt-200 (single-scale)	mask AP	44.21	# 31
Instance Segmentation	COCO minival	ResNeSt-101 (single-scale)	mask AP	41.56	# 38
Instance Segmentation	COCO test-dev	ResNeSt-200 (multi-scale)	AP50	70.2	# 6
Instance Segmentation	COCO test-dev	ResNeSt-200 (multi-scale)	AP75	51.5	# 6
Instance Segmentation	COCO test-dev	ResNeSt-200 (multi-scale)	APS	30.0	# 5
Instance Segmentation	COCO test-dev	ResNeSt-200 (multi-scale)	APM	49.6	# 5
Instance Segmentation	COCO test-dev	ResNeSt-200 (multi-scale)	APL	60.6	# 6
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	box AP	53.3	# 37
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	AP50	72.0	# 19
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	AP75	58.0	# 25
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	APS	35.1	# 22
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	APM	56.2	# 24
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	APL	66.8	# 17
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	Hardware Burden	None	# 1
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	Operations per network pass	None	# 1
Instance Segmentation	COCO test-dev	ResNeSt101	mask AP	43%	# 20
Semantic Segmentation	DADA-seg	ResNeSt (ResNeSt-101)	mIoU	19.99	# 21
Image Classification	ImageNet	ResNeSt-200	Top 1 Accuracy	83.9%	# 181
Image Classification	ImageNet	ResNeSt-101	Top 1 Accuracy	83.0%	# 234
Image Classification	ImageNet	ResNeSt-269	Top 1 Accuracy	84.5%	# 151
Image Classification	ImageNet	ResNeSt-269	Hardware Burden	None	# 1
Image Classification	ImageNet	ResNeSt-269	Operations per network pass	None	# 1
Semantic Segmentation	PASCAL Context	ResNeSt-200	mIoU	58.4	# 11
Semantic Segmentation	PASCAL Context	ResNeSt-269	mIoU	58.9	# 10
Semantic Segmentation	PASCAL Context	ResNeSt-101	mIoU	56.5	# 14

Badge	Markdown
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/instance-segmentation-on-coco)](https://paperswithcode.com/sota/instance-segmentation-on-coco?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/panoptic-segmentation-on-coco-minival)](https://paperswithcode.com/sota/panoptic-segmentation-on-coco-minival?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/semantic-segmentation-on-cityscapes)](https://paperswithcode.com/sota/semantic-segmentation-on-cityscapes?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/semantic-segmentation-on-pascal-context)](https://paperswithcode.com/sota/semantic-segmentation-on-pascal-context?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/semantic-segmentation-on-cityscapes-val)](https://paperswithcode.com/sota/semantic-segmentation-on-cityscapes-val?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/instance-segmentation-on-coco-minival)](https://paperswithcode.com/sota/instance-segmentation-on-coco-minival?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/semantic-segmentation-on-dada-seg)](https://paperswithcode.com/sota/semantic-segmentation-on-dada-seg?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/object-detection-on-coco-minival)](https://paperswithcode.com/sota/object-detection-on-coco-minival?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/object-detection-on-coco)](https://paperswithcode.com/sota/object-detection-on-coco?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/semantic-segmentation-on-ade20k-val)](https://paperswithcode.com/sota/semantic-segmentation-on-ade20k-val?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/semantic-segmentation-on-ade20k)](https://paperswithcode.com/sota/semantic-segmentation-on-ade20k?p=resnest-split-attention-networks)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/resnest-split-attention-networks/image-classification-on-imagenet)](https://paperswithcode.com/sota/image-classification-on-imagenet?p=resnest-split-attention-networks)`

ResNeSt: Split-Attention Networks

19 Apr 2020 · Hang Zhang, Chongruo wu, Zhongyue Zhang, Yi Zhu, Haibin Lin, Zhi Zhang, Yue Sun, Tong He, Jonas Mueller, R. Manmatha, Mu Li, Alexander Smola ·

It is well known that featuremap attention and multi-path representation are important for visual recognition. In this paper, we present a modularized architecture, which applies the channel-wise attention on different network branches to leverage their success in capturing cross-feature interactions and learning diverse representations. Our design results in a simple and unified computation block, which can be parameterized using only a few variables. Our model, named ResNeSt, outperforms EfficientNet in accuracy and latency trade-off on image classification. In addition, ResNeSt has achieved superior transfer learning results on several public benchmarks serving as the backbone, and has been adopted by the winning entries of COCO-LVIS challenge. The source code for complete system and pretrained models are publicly available.

PDF Abstract

Code

Add Remove Mark official

zhanghang1989/ResNeSt official

↳ Quickstart in

Colab

PyTorch Hub

3,008

open-mmlab/mmdetection

19,840

rwightman/pytorch-image-models

↳ Quickstart in

Colab

18,696

dmlc/gluon-cv

5,201

PaddlePaddle/PaddleClas

3,502

See all 27 implementations

Tasks

Add Remove

Image Classification

Instance Segmentation

Object Detection

Panoptic Segmentation

Semantic Segmentation

Transfer Learning

Datasets

ImageNet

COCO

Cityscapes

ADE20K

PASCAL Context

DADA-seg

Results from the Paper

Edit

Ranked #5 on Instance Segmentation on COCO test-dev (APS metric)

Get a GitHub badge

Task	Dataset	Model	Metric Name	Metric Value	Global Rank	Benchmark
Semantic Segmentation	ADE20K	ResNeSt-269	Validation mIoU	47.60	# 54	Compare
Semantic Segmentation	ADE20K	ResNeSt-101	Validation mIoU	46.91	# 58	Compare
Semantic Segmentation	ADE20K	ResNeSt-200	Validation mIoU	48.36	# 45	Compare
Semantic Segmentation	ADE20K val	ResNeSt-101	mIoU	46.91	# 45	Compare
Semantic Segmentation	ADE20K val	ResNeSt-200	mIoU	48.36	# 40	Compare
Semantic Segmentation	ADE20K val	ResNeSt-269	mIoU	47.60	# 42	Compare
Semantic Segmentation	Cityscapes test	ResNeSt200	Mean IoU (class)	83.3	# 9	Compare
Semantic Segmentation	Cityscapes val	ResNeSt-200	mIoU	82.7	# 11	Compare
Object Detection	COCO minival	ResNeSt-200 (single-scale)	box AP	50.54	# 38	Compare
			AP50	68.78	# 17	Compare
			AP75	55.17	# 12	Compare
			APM	54.2	# 13	Compare
			APL	63.9	# 13	Compare
Instance Segmentation	COCO minival	ResNeSt-200 (multi-scale)	mask AP	46.25	# 19	Compare
Object Detection	COCO minival	ResNeSt-200-DCN (single-scale)	box AP	50.91	# 36	Compare
			AP50	69.53	# 14	Compare
			AP75	55.40	# 10	Compare
			APS	32.67	# 12	Compare
			APM	54.66	# 12	Compare
			APL	65.83	# 11	Compare
Instance Segmentation	COCO minival	ResNeSt-200-DCN (single-scale)	mask AP	44.5	# 27	Compare
Panoptic Segmentation	COCO minival	PanopticFPN+ResNeSt(single-scale)	PQ	47.9	# 6	Compare
			PQth	55.1	# 7	Compare
			PQst	37.0	# 6	Compare
Object Detection	COCO minival	ResNeSt-200 (multi-scale)	box AP	52.47	# 31	Compare
			AP50	71.00	# 8	Compare
			AP75	57.07	# 7	Compare
			APS	36.80	# 8	Compare
			APM	56.36	# 9	Compare
			APL	66.29	# 10	Compare
Instance Segmentation	COCO minival	ResNeSt-200 (single-scale)	mask AP	44.21	# 31	Compare
Instance Segmentation	COCO minival	ResNeSt-101 (single-scale)	mask AP	41.56	# 38	Compare
Instance Segmentation	COCO test-dev	ResNeSt-200 (multi-scale)	AP50	70.2	# 6	Compare
			AP75	51.5	# 6	Compare
			APS	30.0	# 5	Compare
			APM	49.6	# 5	Compare
			APL	60.6	# 6	Compare
Object Detection	COCO test-dev	ResNeSt-200 (multi-scale)	box AP	53.3	# 37	Compare
			AP50	72.0	# 19	Compare
			AP75	58.0	# 25	Compare
			APS	35.1	# 22	Compare
			APM	56.2	# 24	Compare
			APL	66.8	# 17	Compare
			Hardware Burden	None	# 1	Compare
			Operations per network pass	None	# 1	Compare
Instance Segmentation	COCO test-dev	ResNeSt101	mask AP	43%	# 20	Compare
Semantic Segmentation	DADA-seg	ResNeSt (ResNeSt-101)	mIoU	19.99	# 21	Compare
Image Classification	ImageNet	ResNeSt-200	Top 1 Accuracy	83.9%	# 181	Compare
Image Classification	ImageNet	ResNeSt-101	Top 1 Accuracy	83.0%	# 234	Compare
Image Classification	ImageNet	ResNeSt-269	Top 1 Accuracy	84.5%	# 151	Compare
			Hardware Burden	None	# 1	Compare
			Operations per network pass	None	# 1	Compare
Semantic Segmentation	PASCAL Context	ResNeSt-200	mIoU	58.4	# 11	Compare
Semantic Segmentation	PASCAL Context	ResNeSt-269	mIoU	58.9	# 10	Compare
Semantic Segmentation	PASCAL Context	ResNeSt-101	mIoU	56.5	# 14	Compare

Methods

Add Remove

1x1 Convolution • ASPP • AutoAugment • Average Pooling • Batch Normalization • Channel-wise Soft Attention • ColorJitter • Convolution • Cosine Annealing • Cutout • DeepLabv3 • Deformable Convolution • Dense Connections • Depthwise Convolution • Depthwise Separable Convolution • Dilated Convolution • DropBlock • Dropout • EfficientNet • FPN • Global Average Pooling • Image Scale Augmentation • Inverted Residual Block • Kaiming Initialization • Label Smoothing • Linear Warmup With Cosine Annealing • LSTM • Mixup • Pointwise Convolution • Random Horizontal Flip • Random Resized Crop • ReLU • Residual Connection • ResNeSt • RMSProp • RoIAlign • SGD with Momentum • Sigmoid Activation • Softmax • Spatial Pyramid Pooling • Split Attention • Squeeze-and-Excitation Block • Swish • Tanh Activation • Weight Decay

Edit Social Preview

ResNeSt: Split-Attention Networks

Code Edit Add Remove Mark official

Tasks Edit Add Remove

Datasets Edit

Results from the Paper Edit

Methods Edit Add Remove

Code

Add Remove Mark official

Tasks

Add Remove

Datasets

Results from the Paper

Edit

Methods

Add Remove