xidaduo2/dsrl

DSRL: Dual Super-Resolution Learning for Semantic Segmentation

The Code is modified from EdgeNet in pytorch, you can follow the procedure in it to prepare the datasets and model directory files.

Testing

• The first step aims to save the gray prediction mask
• The second step aims to evaluate the mIoU with prediction mask and groundtruth

# To evaluate ESPNetv2_DSRL, use below command:
sh run_eval_256x512.sh   
# sh run_eval_512x1024.sh

Main results

Method	s	Image Size	FLOPs	Params	mIOU (class-wise)	Link
ESPNetv2	2. 0	512x256	674.78M	0.79M	54.83% (val)	N/A
ESPNetv2 + DSRL	2.0	512x256	674.78M	0.79M	60.61% (val)	here
ESPNetv2	2. 0	1024x512	2.7G	0.79M	64.44 (val)	N/A
ESPNetv2 + DSRL	2.0	1024x512	2.7G	0.79M	66.50% (val)	here

Citation

If you find this repository helpful, please feel free to cite below work:

@InProceedings{Wang_2020_CVPR,
author = {Wang, Li and Li, Dong and Zhu, Yousong and Tian, Lu and Shan, Yi},
title = {Dual Super-Resolution Learning for Semantic Segmentation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}