Trident Dehazing Network

NTIRE 2020 NonHomogeneous Dehazing Challenge (CVPR Workshop 2020) 1st Solution.

[Challenge report] [TDN paper]

Environment:

• Ubuntu16.04
• Python3.6
• NVIDIA GPU+CUDA8

Dependencies:

• pretrainedmodels==0.7.4
• torchvision==0.2.1
• torch==0.4.1
• tqdm

Test

Compile the DCN module fisrt. If your environment is the same as ours, compile was done. If your pytorch version is 1.0.0, use DCNv2_pytorch1.

Check the hazy images path (test.py line 14), the model path (test.py line 13) and the output path (test.py line 15)

python test.py

Pretrained model

https://pan.baidu.com/s/1l0-hOnIAAbFzmauUmFaRjw password: 22so

https://drive.google.com/file/d/1LcSsCWGLkjmq5o08yhMbSU6DjCGugmRw

Another solution (2nd Solution, Knowledge Transfer Dehazing Network) of our team:

https://github.com/GlassyWu/KTDN