a real-time lightweight network for semantic segmentation in rainy environments, shows high performance compared with other lightweight semantic segmentation networks(BiSeNet、CGNet、ContextNet、DFANet、EDANet、ENET、ERFNet、ESPNet、FDDWNet、LEDNet、LinkNet、UNet（add none-local、use heavy backbones（resnet、resnext、SEresnet）、change the structures)、PSPNet、SEGNet、UNet_nest.
add lovaze-softmax loss\focal loss for segmentation
environments: Ubuntu18.04 torch 1.1.0 CUDA10.1
For train： sudo python3 --action="train" --choose_net="NET" NET: is the parameter you can choose in main.py file, you also can change the parameters directly in main.py file. for example:sudo python3 main.py --action="train" --choose_net="my_drsnet_A"
For test: sudo python3 --action="test" --choose_net="NET" for example:sudo python3 main.py --action="test" --choose_net="my_drsnet_A"
to change datasets: first,download the add-rain datasets from the link below, and change the folder name of dataset in main.py
to get original UAS dataset:https://pan.baidu.com/s/1IWSVKYBrYwxaRThPfDsDGg to get original BPS dataset:http://www.cbsr.ia.ac.cn/users/ynyu/dataset/
to get UAS-add-rain and BPS-add-rain datasets: link:https://pan.baidu.com/s/1zRcBd2vTuKFWTIsQJrbQnA password: awfy
to get pretrained weights of different scales on UAS-ad-rain and BPS-ad-rain datasets:
link: https://pan.baidu.com/s/1Xv98jfHIJeyToV3PwfEJfQ
password: 0tt9
the article related to this code is available here:https://link.springer.com/article/10.1007/s11554-020-01042-2?wt_mc=Internal.Event.1.SEM.ArticleAuthorOnlineFirst&utm_source=ArticleAuthorOnlineFirst&utm_medium=email&utm_content=AA_en_06082018&ArticleAuthorOnlineFirst_20201103
any problems, please do not hesitate to contact with me, my email is [email protected].