CameraNet-Tensorflow

CameraNet: A Two-Stage Framework for Effective Camera ISP Learning

[[Paper](http://www4.comp.polyu.edu.hk/~cslzhang/paper/CameraNet.pdf]]

Environment

• Tensorflow == 1.15.0
• Cuda == 10.1
• Python == 3.7

Datasets

• HDR+
• SID

Usage

Currently the code is ugly. We will try to simplify the code or add comments in the future for better reading.

For HDR+ ISP

1. Make a dataset directory in the root folder by mkdir Data.
2. Download the HDR+ datasets (already including training and testing sets). Unzip it to Data folder. Now you should have a folder named ./Data/HDRp
3. For training, python train_hdrp.py
4. For testing, python test_hdrp.py

For SID ISP

1. Make a dataset directory in the root folder by mkdir Data.
2. Download the SID datasets (already including training and testing sets). Unzip it to Data folder. Now you should have a folder named ./Data/SID
3. For training, python train_sid.py
4. For testing, python test_sid.py Note that for SID, in the paper we use a different training-testing separation of the data from the original SID paper. To allow a beter comparison, in this code we adopt the training-testing separation from the original SID paper. The PSNR and SSIM are a little different from our paper but remain in the same level.

Contact

Zhetong Liang zhetong.liang@connect.polyu.hk

Citation

@ARTICLE{9329084, author={Liang, Zhetong and Cai, Jianrui and Cao, Zisheng and Zhang, Lei}, journal={IEEE Transactions on Image Processing}, title={CameraNet: A Two-Stage Framework for Effective Camera ISP Learning}, year={2021}, volume={30}, number={}, pages={2248-2262}, doi={10.1109/TIP.2021.3051486}}