Course project of CMU 15-663
In this project, we aim to implement a learning-based method for recovering images captured in the dark. Using a Unet-like architecture, we try to train our model to perform the image processing pipeline in an end ot end fashion.
- Python 3.6.0
- rawpy
- PyTorch 1.0
python train_Sony.py --gpu 0
python test_Sony.py --model ./model.pl --gpu 0
python metrics.py --imgdir ./eval
PSNR | SSIM | |
---|---|---|
NLM | 16.212 | 0.225 |
LSID | 28.575 | 0.815 |
Ours | 28.625 | 0.818 |
MIT license