This code is outdated and no longer maintained, please use https://github.com/mgharbi/demosaicnet instead
Runs on python3 (tested on Anaconda for Ubuntu 14.03). Other dependencies are listed in requirements.
- Install and download the data
make setup
- Launch a visdom server to monitor training and display a few images
make server
- Train a net
make train_demo
- Test a net (todo)
The dataloader assumes the dataset is given as a listing file containing the relative path to the images. For example if you have images like:
root
├── filelist.txt
├── hdrvdp
│ ├── 000
│ │ ├── 000001.png
│ │ ├── 000002.png
│ │ ├── 000003.png
│ │ ├── 000004.png
│ │ ├── 000005.png
filelist.txt should have one path per line as such:
hdrvdp/000/000001.png
hdrvdp/000/000002.png
hdrvdp/000/000003.png
hdrvdp/000/000004.png
hdrvdp/000/000005.png
etc.