X-Trans demosaic algorithms differences

Started 4 months ago | Discussions thread
saltydogstudios Regular Member • Posts: 232
Re: X-Trans demosaic algorithms differences

sluggy_warrior wrote:

saltydogstudios wrote:

WOW. Thank you so much for posting these samples.

I've gained so much insight from pixel peeping these images. Would you be OK with me using these images in my blog ? I'll credit you however you like.

Thanks for sharing your blog, very interesting to see your finding.

Feel free to use any of my images (RAF and JPG), no need to credit me. The street scene with the Chinese characters sign actually belongs to Marco1971, so please check with him/her on that.

Okay. I may just take my own - I was hoping to avoid the extra work, but if I take the same photo with a Bayer and X-Trans it would be a better apples-to-apples example.

A few typos/grammar I spotted on your blog (I'm NNS, ignore if I'm wrong Also, as others have corrected you that the number of green/red/blue pixels are not the same for Bayer vs X-Trans. Let us know when you have the article revised and polished.

Thank you, I have edited that, hopefully the site has updated by now.

One thing I just found out after looking at darktable output from BryceM: the extra sharpness actually came from my denoise combo. Here's the output of the same signs scene, using the same dt style but with the three "Profiled Denoise" instances turned off.

darktable, Markesteijn 3-pass, 5-times, no denoise

darktable, Markesteijnm 3-pass, 5-times, 3 Profiled Noise instances

So, you might want to look into what those three denoise instances do.

Thanks. I purposefully exaggerated things for the purposes of demonstration, but I'll look into this.

I tweaked them originally for D5500, but mostly for denoising high ISO and it did OK at that. However, the same combo somehow works really well for X-Trans and even sharpens up the details. At this point I have no idea how it works. I understand the wavelets and NLM algorithms at basic level only, but have no idea how they interact with each others, it was mostly trial-and-error on the parameters (and lots of pixel-peeping) over the years.

Maybe Frank Markesteijn hit the jackpot by accident in the same way

FIWI, here's the details of that style:

[Demosaic]
- Markesteijn 3-pass, 5-times smoothing

[Profiled denoise 1]
- wavelets strength=0.5
- blend uniformly, mode=average, opacity=50%

[Profiled denoise 2]
- wavelets strength=1
- blend uniformly, mode=HSV-color, opacity=50%

[Profiled denoise 3]
- NLM patch-size=4 strength=4
- blend uniformly, mode=hue, opacity=85%

 saltydogstudios's gear list:saltydogstudios's gear list
Ricoh GR Digital Ricoh GR Digital III Sigma DP2 Merrill Sigma DP3 Merrill Fujifilm X-Pro1 +3 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow