Awesome

dehaze_release

This is the PyTorch code for ''Strong Baseline for Single Image Dehazing with Deep Features and Instance Normalization'' publisehd in BMVC 2018. The arxiv version is here.

The pre-trained model can be found here.

To test the pre-trained model, put the downloaded models in folder named ''models'', put the RESIDE standard in ''data'', and run

python main.py --trans-flag in --use-bn in  --test-flag --test-batch-size 8 --gpuid 0 --load-model models/dehaze_release.pth --save-image output

The dehazed images can be found in folder ''output''. The pre-trained model could achieve PSNR 27.79 and SSIM 0.9556 on RESIDE_standard, evaluated by the matlab script provided on the dataset webpage.

To train a model, please run

python main.py --trans-flag in --use-bn in --batch-size 16 --test-batch-size 8 --optm sgd --lr 0.1 --lr-freq 30 --epochs 60 --rec-w 1 --per-w 1  --print-freq 200 --gpuid 0,1,2,3

citation

@article{xu2018effectiveness, title={Strong Baseline for Single Image Dehazing with Deep Features and Instance Normalization}, author={Xu, Zheng and Yang, Xitong and Li, Xue and Sun, Xiaoshuai}, journal={BMVC}, year={2018} }

acknowledgement

We thank the released PyTorch code and model of WCT style transfer.