Skip to content

Latest commit

 

History

History

isnet

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

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%/81.31% 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%/81.33% 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%.