CSANet for Video Semantic Segmentation With Inter-Frame Mutual Learning (IEEE Signal Processing Letters 2021)

Enviroment

• Titan RTX(24G) * 8
• pytorch(1.7.1+)

Models

ResNet101 Pretrained model：https://pan.baidu.com/s/1Lwzd8dDN12OCQbif64TIrg (IIAU)

Best.pth is available: https://pan.baidu.com/s/1Mx_rn1k_jEhNOOnhDQqKtQ (IIAU)

Training

sh train.sh

Validation

sh val.sh

Testing(submit)

sh test.sh

Results on Cityscapes val:

CSANet lr-decay-99600 79.16%

CSANet lr-decay-99200---1e-4-finetune-111600 79.61%

CSANet 111600-ms 80.76%

Acknowledgment

OCNet