Le 28/04/2014 00:48, Tobias Ellinghaus a écrit : >> Subject says it all. I assumed (perharps incorrectly) that the denoising >> would bw one of the last things to process in the pipeline. In my darktable >> copy it appears right at the beginning (see screenshot). It this correct? > Actually it should be even earlier in the pipe. That's something that will > likely change in the not too distant future. I don't see why it would be > beneficial to move it to the end.
Please complete/correct if I'm wrong. # Early profiled denoising is good Profiled denoising is a special case. Since it is based on noise profile specifically measured by sensor, ISO rating and local luminance, it must be before any other module that may change local luminance, or it would lose relevance. Yet it's just after demosaicing module which drastically alters the data format. So putting it earlier needs changing it. It may be relevant (after all, profiled denoising is, well, profiled for a specific sensor so we can assume sensor-specific properties like mosaic geometry and the like). That is difficult and would increase module complexity. # Other denoising should also happen before Other denoising modules (non-local, bilateral) do happen later. I would expect them to happen not so late, for a different reason : noise being a per-pixel value, it makes sense to deal with it before any module has a chance to displace pixels, like rotation, lens correction, perspective. Lens correction may not look too troublesome in case of small corrections, which is the case most of the time. But some case like when the output geometry (rectilinear, equirectangular, fisheye, etc) is different from the lens geometry make high local distortion which probably hinders noise algorithm. Similarly, the "crop and rotate" module, especially when rotation or keystone is enabled for perspective correction, highly change the local pixel geometry. Denoising modules generally assume a local geometry or some sort of kernel. The more the local geometry is distorded, the more the denoising operates on data that mismatches assumptions. This suggests that denoising modules should be done before lens correction. Yet, other filters like base curve and tone curve drastically change luminance distribution in the picture. I can explain why I think these should happen after exposure module, but should those denoising happen before or after base curve and tone curve ? Thank you for your attention. Comments welcome. -- Stéphane Gourichon ------------------------------------------------------------------------------ "Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE Instantly run your Selenium tests across 300+ browser/OS combos. Get unparalleled scalability from the best Selenium testing platform available. Simple to use. Nothing to install. Get started now for free." http://p.sf.net/sfu/SauceLabs _______________________________________________ Darktable-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/darktable-users
