idrnet

Introduction

Official Repo

Code Snippet

IDRNet (NeurIPS'2023)

@inproceedings{jin2023idrnet,
    title={IDRNet: Intervention-Driven Relation Network for Semantic Segmentation},
    author={Jin, Zhenchao and Hu, Xiaowei and Zhu, Lingting and Song, Luchuan and Yuan, Li and Yu, Lequan},
    booktitle={Thirty-Seventh Conference on Neural Information Processing Systems},
    year={2023}
}

Results of Different Frameworks

LIP

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU	Download
FCN	ImageNet-1k-224x224	R-50-D8	473x473	LR/POLICY/BS/EPOCH: 0.01/poly/32/150	train/val	51.24%	cfg | model | log
PSNet	ImageNet-1k-224x224	R-50-D8	473x473	LR/POLICY/BS/EPOCH: 0.01/poly/32/150	train/val	53.29%	cfg | model | log
UperNet	ImageNet-1k-224x224	R-50-D8	473x473	LR/POLICY/BS/EPOCH: 0.01/poly/32/150	train/val	54.00%	cfg | model | log
DeepLabV3	ImageNet-1k-224x224	R-50-D8	473x473	LR/POLICY/BS/EPOCH: 0.01/poly/32/150	train/val	53.87%	cfg | model | log

ADE20k

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU	Download
FCN	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.01/poly/16/130	train/val	43.61%	cfg | model | log
PSNet	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.01/poly/16/130	train/val	44.02%	cfg | model | log
UperNet	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.01/poly/16/130	train/val	44.84%	cfg | model | log
DeepLabV3	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.01/poly/16/130	train/val	44.75%	cfg | model | log

CityScapes

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU	Download
FCN	ImageNet-1k-224x224	R-50-D8	512x1024	LR/POLICY/BS/EPOCH: 0.01/poly/8/220	train/val	79.91%	cfg | model | log
PSNet	ImageNet-1k-224x224	R-50-D8	512x1024	LR/POLICY/BS/EPOCH: 0.01/poly/8/220	train/val	79.88%	cfg | model | log
UperNet	ImageNet-1k-224x224	R-50-D8	512x1024	LR/POLICY/BS/EPOCH: 0.01/poly/8/220	train/val	80.81%	cfg | model | log
DeepLabV3	ImageNet-1k-224x224	R-50-D8	512x1024	LR/POLICY/BS/EPOCH: 0.01/poly/8/220	train/val	80.69%	cfg | model | log

COCOStuff-10k

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU	Download
FCN	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.001/poly/16/110	train/test	38.61%	cfg | model | log
PSNet	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.001/poly/16/110	train/test	39.13%	cfg | model | log
UperNet	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.001/poly/16/110	train/test	39.35%	cfg | model | log
DeepLabV3	ImageNet-1k-224x224	R-50-D8	512x512	LR/POLICY/BS/EPOCH: 0.001/poly/16/110	train/test	39.31%	cfg | model | log

SOTA Results

PASCAL-Context-59

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU/mIoU (ms+flip)	Download
UperNet	ImageNet-22k-384x384	Swin-Large	640x640	LR/POLICY/BS/EPOCH: 0.00006/poly/16/260	train/val	63.82%/64.50%	cfg | model | log

LIP

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU/mIoU (flip)/mIoU (ms+flip)	Download
UperNet	ImageNet-22k-384x384	Swin-Large	473x473	LR/POLICY/BS/EPOCH: 0.00006/poly/16/110	train/val	60.53%/60.83%/61.17%	cfg | model | log

ADE20k

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU/mIoU (ms+flip)	Download
UperNet	ImageNet-22k-384x384	Swin-Large	640x640	LR/POLICY/BS/EPOCH: 0.00006/poly/16/130	train/val	53.97%/54.68%	cfg | model | log

COCOStuff-10k

Segmentor	Pretrain	Backbone	Crop Size	Schedule	Train/Eval Set	mIoU/mIoU (ms+flip)	Download
UperNet	ImageNet-22k-384x384	Swin-Large	640x640	LR/POLICY/BS/EPOCH: 0.00006/poly/16/110	train/test	49.94%/50.54%	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

Please note that, due to differences in computational precision, the numerical values obtained when testing model performance on different versions of PyTorch or graphics cards may vary slightly. This is a normal phenomenon and the performance differences are generally within 0.1%.