PyTorch-Image-Dehazing
PyTorch implementation of some single image dehazing networks.
Currently Implemented: AOD-Net: An extremely lightweight model (< 10 KB). Results are good.
Prerequisites:
- Python 3
- Pytorch 0.4
Preparation:
- Create folder "data".
- Download and extract the dataset into "data" from the original author's project page. (https://sites.google.com/site/boyilics/website-builder/project-page).
Training:
- Run train.py. The script will automatically dump some validation results into the "samples" folder after every epoch. The model snapshots are dumped in the "snapshots" folder.
Testing:
- Run dehaze.py. The script takes images in the "test_images" folder and dumps the dehazed images into the "results" folder. A pre-trained snapshot has been provided in the snapshots folder.
Evaluation:
WIP.
Some qualitative results are shown below:
