This is my Pytorch implementation of Learning to See in the Dark in CVPR 2018. For more information, please read the Report.pdf and See-In-The-Dark notebook.
Original paper by Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun.
Original Tensorflow implementation by Chen Chen
| PSNR on test set | No. Epochs | |
|---|---|---|
| Chen et al. | 28.88 | 4000 |
| My implementation | 28.83 | 4000 |
Training PSNR and L1 Loss
Denoise a random image in original dataset
Denoise a Sony A7sii raw image found on Google
Testing cross sensor generalizaion
This image was taken on a Sony RX100VI with a 1-inch Bayer sensor
- 64GB RAM
- Pytorch 1.0
- Numpy + Rawpy
- Matplotlib
I ran the code on a Tesla V100 32GB. If your GPU has less memory, reduce the batch size.