Skip to content

Latest commit

 

History

History

isnet

Folders and files

NameName
Last commit message
Last commit date

parent directory

..

README.md

README.md

 
 

README.md

Introduction

Official Repo

Code Snippet

ISNet (ICCV'2021) 
@inproceedings{jin2021isnet,
    title={ISNet: Integrate Image-Level and Semantic-Level Context for Semantic Segmentation},
    author={Jin, Zhenchao and Liu, Bin and Chu, Qi and Yu, Nenghai},
    booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
    pages={7189--7198},
    year={2021}
}

Results

COCOStuff-10k

Backbone Pretrain Crop Size Schedule Train/Eval Set mIoU/mIoU(ms+flip) Download
R-50-D8 ImageNet-1k-224x224 512x512 LR/POLICY/BS/EPOCH: 0.001/poly/16/110 train/test 38.06%/40.39% cfg | model | log
R-101-D8 ImageNet-1k-224x224 512x512 LR/POLICY/BS/EPOCH: 0.001/poly/16/110 train/test 40.53%/41.74% cfg | model | log
S-101-D8 ImageNet-1k-224x224 512x512 LR/POLICY/BS/EPOCH: 0.001/poly/32/150 train/test 41.55%/42.53% cfg | model | log

ADE20k

Backbone Pretrain Crop Size Schedule Train/Eval Set mIoU/mIoU(ms+flip) Download
R-50-D8 ImageNet-1k-224x224 512x512 LR/POLICY/BS/EPOCH: 0.01/poly/16/130 train/val 44.22%/45.03% cfg | model | log
R-101-D8 ImageNet-1k-224x224 512x512 LR/POLICY/BS/EPOCH: 0.01/poly/16/130 train/val 45.92%/47.29% cfg | model | log
S-101-D8 ImageNet-1k-224x224 512x512 LR/POLICY/BS/EPOCH: 0.004/poly/16/180 train/val 46.65%/47.56% cfg | model | log

CityScapes

Backbone Pretrain Crop Size Schedule Train/Eval Set mIoU/mIoU(ms+flip) Download
R-50-D8 ImageNet-1k-224x224 512x1024 LR/POLICY/BS/EPOCH: 0.01/poly/16/440 train/val 79.32%/80.92% cfg | model | log
R-101-D8 ImageNet-1k-224x224 512x1024 LR/POLICY/BS/EPOCH: 0.01/poly/16/440 train/val 80.56%/81.96% cfg | model | log
S-101-D8 ImageNet-1k-224x224 512x1024 LR/POLICY/BS/EPOCH: 0.01/poly/16/440 train/val 78.78%/81.30% cfg | model | log

LIP

Backbone Pretrain Crop Size Schedule Train/Eval Set mIoU/mIoU(flip) Download
R-50-D8 ImageNet-1k-224x224 473x473 LR/POLICY/BS/EPOCH: 0.01/poly/32/150 train/val 53.14%/53.41% cfg | model | log
R-101-D8 ImageNet-1k-224x224 473x473 LR/POLICY/BS/EPOCH: 0.01/poly/32/150 train/val 54.96%/55.41% cfg | model | log
S-101-D8 ImageNet-1k-224x224 473x473 LR/POLICY/BS/EPOCH: 0.007/poly/40/150 train/val 56.52%/56.81% cfg | model | log

More

You can also download the model weights from following sources:

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%.