Whatever the DXO One camera with Super Raw mode does, give it to us in PhotoLab where we could choose multiple images of the same scene, and the software would reduce noise, increase resolution and demosaic using AI, like DeepPrime XD does but I believe ML models would benefit from multiple images.
We have seen some algorithms which reduce the motion blur of moving subjects and merge multiple images into one image from mobile phones such as iPhone and Google Pixel, the same thing would be awesome to have for raw files, or any kind of format for that matter.
Maybe a good starting point to research and implement:
It may be asking for the same result, but the approach is different. The approach I am proposing is to use the latest AI models to fuse the images together instead of relying on hand made tailored algorithms.
But whatever DXO team decides on, I am willing to test it and be part of it.
It would be great to support all of those different modalities, but lets keep it one step at a time.
Merging HDR and focus are completely different things in my mind.
Astro would be a fun thing to do if the motion of the stars can be tamed by AI, keeping the stars sharp and the foreground as well if the camera was not using motorized stars tracker thingy. I know that software like Sequator does a good job of stacking astro photos.
It is an open-source and free tool that can merge a burst of images captured handheld. The merging is robust against camera shake and scene motion (to a certain degree) and the resulting image has reduced noise. The interesting part is that the tool acts directly on RAWs and then produces a Raw DNG file that behaves very similar to a RAW file. You can e.g. apply Deep Prime and lens corrections to the file in Photolab and follow your typical workflow. The current version published in the App Store is not yet fully compatible with Photolab (e.g. lens sharpness does not work) and the merging approach is relatively simple. A beta version (not yet published) improves compatibility and has improved motion robustness and noise reduction capability. As a prerequisite, the free tool Adobe DNG converter has to be pre-installed. Please find examples and instructions on the website posted above.
Wow, thanks for the info!
I might and try to convert this code into something more portable if I find the time.
I do not have a Mac. It would be great if this app could run on all platforms.
I fully agree that a platform-independent solution would be even better. The aim of the project is to reproduce the Google algorithm HDR+, which is the basis for Google night sight:
Currently, the Fourier-based merging approach is not yet published, but it already works and is currently being tested.