Back to forums

RawNerf - an emerging low light processing technology

Processing Noise Reduction Deep Sky
3 replies•223 views
David Wright avatar
David Wright avatar
First time I've posted to the combined intelligence of the Astro community.  I just ran across this new technology and was wondering whether anyone had seen it and whether there is an astro application?

Looks like a really interesting noise-reduction technology for low-light imaging.  

https://bmild.github.io/rawnerf/

Hope you find this interesting.

Conflict note:  I have no interest in this beyond being a beginning astro nerd....
Gernot Schreider avatar
Gernot Schreider avatar
Hi @David Wright 

I saw the anouncement by Google as well. It is a proof of concept so far. They published the code on github and you are free to use it. You will have to compile it with a bunch of libraries and other stuff. However there is no application yet that makes it usable to the wider public.
Seems that we need to wait some time until this becomes available, if at all.

The demo-video of the capabilities is indeed impressive.

Cheers
Gernot
Helpful
kuechlew avatar
kuechlew avatar
At 1:25 of the video: "our method is able to combine images taken from many different camera viewpoints to jointly denoise and reconstruct the scene."
Good luck taking astro photos from many different camera viewpoints …

Clear skies
Wolfgang
Benoit Blanco avatar
Benoit Blanco avatar
Hi @David Wright,

The problem for an astro application is that it's a trained algorithm. From my experience, any trained algorithm can shows some "hallucinations" ; things that doesn't exist in the picture but which was present in the training dataset.

Moreover, this technique works by stacking different views of the same object from different angles. That is not possible from earth.. Even if we wait 6 months between 2 pictures, the angle will be too small for a nebula. 

I found deepprior to be a good way de denoise astro images with deep learning. It removes random noise without training. It basically trains itself to recreate the input image, and of course random noise can't be restored. It's a heavy process and require large GPU but I used it in several of my processings.
Helpful Insightful Respectful