aiff22/PyNET-PyTorch

A comparsion with the AIM2020 Image Signal Processing Challenge

jasonchenPJ opened this issue · 2 comments

jasonchenPJ commented

The Pynet use the test raw(448x448) of ZRR dataset to evaluate the PSNR result. Are the results comparable to AIM 2020 ISP pipeline challenge results ?
image

aiff22 commented

Hi @jasonchenPJ,

No, in the AIM 2020 challenge we used additional test images that are not available publicly.

YangGangZhiQi commented

@aiff22 Hi, thanks for your great work. When using your model to inference my local pics with the order of (ch_B, ch_Gb, ch_R, ch_Gr), I get a red rendered results. It seems like there are big difference between the same bayer pattern,eg same [[R,Gr,R, Gr],[Gb,B,Gb,B]]。Is there a solution to transfer local pic to your bayer pattern so that your model could predict correctly.

  • As I known, in 'ISP with deep learning' field, the training dataset are hard to collect. The model with public dataset may not be available to local pictures. I think the transfer method above have great value for practice use or research. Hope for your reply.