Gauging Noise Reduction Performance...Why not try???

Just an FYI for any FOSS dev or user that might have an interest in noise reduction. It might be interesting to send in a couple of images and see what these guys have come up with. I know there are are also products from DXO and Topaz but ON1 is doing a free offer to process your image.

You could compare it back against what we can achieve with the available FOSS software just to see.

Just throwing it out there as more of an exercise than any sort of endorsement or promotion…

https://www.on1.com/products/nonoise-ai/nonoise-challenge/?utm_campaign=Engage_NoNoiseChallenge&utm_source=RAW21_Own&utm_medium=email&popup=no&mkt_tok=MTg1LVJXWC02MTgAAAF89v7f-yx_zN5QDDZI-U_YIim4H35zAVdUTRKPC3qrebnXNoCPGeu_hhACmzksyi43ZeMzyGCSI7C167eivcmkbIyI_K4PtsCneF2TFQ

The cynic in me says they want people to help them train their machine learning algorithm. :imp:

2 Likes

Likely…it could be like playing chess against the computer thought…it is pretty funny many of UI upgrades in their last edition were strait from the feature list in DT…

Taking from FLOSS is as old as time (well, computers). That is why it is a good philosophy.

2 Likes

Wouldn’t they need some kind of ‘ground truth’ for that? They just ask for noisy images. But for training of their AI they could scrape CC0 images for that…or am I missing something?

Hm, maybe the preselection of ‘give me your best but noisiest images’ helps them build their noise estimation precision for non-test-images? Curious!

They want free publicity (which they are getting here).

They want loads of noisy images to test and tweak their software.

They want to show off some before/after versions. I suppose they will choose examples that show their software in the best light.

1 Like

It’s built in to a marketing campaign right so who even knows if they are getting any utility out of it but they can send back or post some nice before and after images and give away a few copies and try to create some buzz……

1 Like

There are different sub-problems to be explored, one being single image blind denoising, and various methodologies with which one could use to measure how successful an enhancement is.