W-Net: Two-Stage U-Net With Misaligned Data for Raw-to-RGB Mapping

Environment

• Ubuntu 16.04 LTS.
• Pytorch 1.2.
• CUDA 10.0.

Run inference on ZRR test images

1. Set argument 'raw_dir' to the directory containing input raw images

2. Download the pretrained models in here, and place them under the folder named "trained_model"

• To obtain the result images of 'fidelity track', run test.py

• To obtain the full resolution result images, run test_fullres.py

Reference

Kwang-Hyun Uhm, Seung-Wook Kim, Seo-Won Ji, Sung-Jin Cho, Jun-Pyo Hong, and Sung-Jea Ko, "W-Net: Two-Stage U-Net With Misaligned Data for Raw-to-RGB Mapping," 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), Seoul, Korea (South), 2019, pp. 3636-3642. link