Closed TomHeaven closed 1 year ago
This work employs masked denoising as an auxiliary task, making it suitable for pixel-wise independent noise. The noise in the raw domain of the SIDD dataset conforms to this requirement. However, in the sRGB domain, demosaicing and filtering operations often degrade the spatial independence of noise in the raw domain, introducing spatially correlated structured noise. Therefore, it is recommended to use Blind2Unblind in the raw domain rather than the sRGB domain.
Thanks for the explanation.
Thanks for the nice paper and code.
I run the code on SSID Raw Validation dataset (ValidationNoisyBlocksRaw.mat) and it works. The generated log file (A_log_SIDD_Val.csv ) looks good as:
However, When I run the code (with slight modifications to fit the three-channel input) on SSID sRGB Validation dataset (ValidationNoisyBlocksSrgb.mat). The generated log file (A_log_SIDD_Val.csv ) shows that the model seems to converge to the identity mapping.
A visual result of SIDD_Val_023-027-005_mid looks like this:![SIDD_Val_023-027-005_mid](https://user-images.githubusercontent.com/11656711/231137019-9a895f5f-6838-4ce0-beba-38e645b7bf1f.png)
Have you tried the code on SSID sRGB Validation dataset? Any suggestions are appreciated.