GCP-Net
GCP-Net copied to clipboard
GuoShi28
# About the SNR of an image
Hello，thanks for your great job！ In this paper，you demonstrated the Green channel SNR of an Bayter image is higher than Red and Blue channel, I wonder if this law exists in sRGB image? By the way , how to compute the SNR of an image in SIDD? Looking forward to your reply !
Thank you. The green channels have higher SNR than other channels. In sRGB space, it's maybe more complicated effected by the ISP process. In SIDD dataset, you can use the noise image and the corresponding ground-truth image to calculate SNR. The SNR function can use snr() in matlab or simple write by yourself.
Hello,in your GCP_Net ,the input should be concat of image and noise map , i want to know if i captured raw burst , how can i get the noise map ? In your code , you provided the raw data and the organized data , can you provide the code of organizing the not processed raw image ?
Looking forward to your reply! Thanks again for your great work！
@GuoShi28
For the raw image, the noise levels are in the metadata of raw data. It exists in raw format. If you can not obtain the noise level, you can directly remove the noise map in the input. I'm not fully understand "can you provide the code of organizing the not processed raw image ?". For the ISP and inverse ISP, the code is provided in the process.py and unprocess.py files.
In your test_real.py file , the raw image has been processed and saved as ".mat" file , and the meta data has been saved in ".mat" file, and the captured raw file is end with "DNG", i wonder how to process the DNG file to the ".mat" file ? Specificly, there are a lot of tags in DNG file , I do not know which one is the noise level .