UnprocessDenoising_PyTorch icon indicating copy to clipboard operation
UnprocessDenoising_PyTorch copied to clipboard

Question about infer for real raw

Open cw-plus opened this issue 1 year ago • 1 comments

Hi, when I use a raw image for inference, I find that the Net must be fed with both the input image and its corresponding input variance. How can I obtain the input variance?

cw-plus avatar May 09 '24 10:05 cw-plus

Hi,

Sorry for the late reply. The method is designed to perform denoising on RAW images provided their noise variances are available in their corresponding metadata files. So, I think without the value of input variance, you may not be able to use this method for your case.

However, you may use a workaround. You can use your input raw image and take input variance from the DND dataset (there are various ways to do this like use the average input noise variance of the entire dataset, or just visually inspect which image in the dataset looks visually similar to your image and then take the input variance of that image). Hope it helps.

aasharma90 avatar Jun 05 '24 06:06 aasharma90