Lightroom Classic: Demosiacing is falling behind and needs improvement

  • 1
  • Idea
  • Updated 1 month ago
  • (Edited)
Hi there,

Maybe I am a pixel peeper, but as of 2020, my observations are that Lightroom's demosaicing is falling way behind competitors and doesn't recover as much detail as, for example, CaptureOne or RawTherapee. Together with fairly average sharpening and denoising, the final result is so mushy compared to what you can get elsewhere with just a snap of your fingers, that it's just frustrating and urges me to look for a replacement.

Here is a 200% crop of a photo taken on Nikon D5100 as demosaiced by RawTherapee's AMaZE VS Lightroom (all sharpening, denoising and corrections turned off in both). See how much crisper and detailed is the AMaZE version?



It's literally free resolution! Well, technically, the Lr's version is "lost resolution" - you are not getting from your gear what you paid for it.

Can Adobe do something about this? I mean, adopting AMaZE demosaicing from the open-source shouldn't be too hard, right? That would be already a significant improvement. Please?

Cheers!
Photo of Sergei Datsenko

Sergei Datsenko

  • 6 Posts
  • 0 Reply Likes
  • disappointed

Posted 1 month ago

  • 1
Photo of Rikk Flohr

Rikk Flohr, Official Rep

  • 6690 Posts
  • 1507 Reply Likes
It might be useful, Sergei to compare an Enhanced Details version as part of your comparison. Have you evaluated this feature?
Photo of Verochka Tzimmermann

Verochka Tzimmermann

  • 2 Posts
  • 1 Reply Like
it is a great feature, which unfortunately forces everybody (even those with RTX2080 GPUs) to create massive, massive (linear+ sized) DNG files instead of letting users with powerful GPUs to do that type demosaick w/o creating intermediate DNGs... I am using on-line backups and to deal with extra huge DNGs is too much... address that first
Photo of Sergei Datsenko

Sergei Datsenko

  • 6 Posts
  • 0 Reply Likes
The difference between non-enhanced and enhanced is very marginal, mostly in fringing artifacts.

Here is enhanced version too:


RawTherapee is still ahead by a lot.

Photo of Jack Nilles

Jack Nilles

  • 24 Posts
  • 5 Reply Likes
Sergei, it may be that the details were lost in the translation but I don't see any difference between the RT and Lr examples you gave. At least on my monitor.
Photo of Sergei Datsenko

Sergei Datsenko

  • 6 Posts
  • 0 Reply Likes
You need to open the image, embedded one is downsized and it's less obvious, though still visible.
Photo of Johan Elzenga

Johan Elzenga, Champion

  • 2538 Posts
  • 1073 Reply Likes
I have to agree with Jack. The left image has a little more contrast which may make it seem a little sharper, but I do not see any difference in detail.
Photo of Sergei Datsenko

Sergei Datsenko

  • 6 Posts
  • 0 Reply Likes
That's the thing, it's not the contrast, if you look at the deep shadows the Lr's version has actually more overall contrast than RT's one. What you refer to is microcontrast, e.g. the ability to maintain hard edges hard and don't smudge them across several pixels, and this is exactly what we want from a good demosaicing algorithm. Lightrooms version feels like a tiny bit of additional gaussian blur on top of the image, and as far as I am aware there is no reliable way to "undo" that.

This actually makes me think that if I find a good example from a camera that doesn't have an AA-filter on the sensor the difference might be even more pronounced.

Photo of Johan Elzenga

Johan Elzenga, Champion

  • 2538 Posts
  • 1073 Reply Likes
Well, whatever we want to call it, I cannot agree with your statement that RT is way ahead. If you pixel-peep enough you may see very slight differences, but that is all I can see. Very slight differences that I am sure will not show at all if you compare the images in a normal way rather than 200% zoomed in on screen, i.e. if you would print them, for example. You are making a big deal out of nothing in my opinion.
Photo of Sergei Datsenko

Sergei Datsenko

  • 6 Posts
  • 0 Reply Likes
Obviously there is some sort of touchiness around mentioning RT here :)

Sure, it's not that a big deal, especially if you are using half a hundred megapixel full-frame camera and have pixels to spare, but this becomes much more of an issue on smaller sensors. And again, the question is about demosaicing algorithms which is inherently a pixel-peeping question and requires 200% or even 400% zooms. But at the end of the day, better inputs will make editing easier and results in some sense better.
Photo of Todd Shaner

Todd Shaner, Champion

  • 1711 Posts
  • 580 Reply Likes
Here is a 200% crop of a photo taken on Nikon D5100 as demosaiced by RawTherapee's AMaZE VS Lightroom (all sharpening, denoising and corrections turned off in both). See how much crisper and detailed is the AMaZE version?
IMHO this is a pointless comparison. All raw files need sharpening and Tone settings applied to the demosaiced image data. What does this image file look like AFTER applying ALL Develop module controls including Detail panel adjustments compared to RawTherapee? In other words your best effort editing the image in both raw converters. What camera model?
Photo of Sergei Datsenko

Sergei Datsenko

  • 6 Posts
  • 0 Reply Likes
What would be the point of such comparison? If I thought that RT is better in every way possible I would just use it and save myself the Lr's subscription fee. I use it as an example because it's really easy to get control over each step there and see what exactly is going on. I could use C1 as an example - it also does better than Lr - but it's not as easy to ensure that nothing else is happening under the hood.

I isolated a single step that lies in the foundation of all following processing, capture sharpening will happen later and it's irrelevant at this stage. Better inputs - better outputs and more control over editing. But if you've lost definition at that very first step - there is nothing you can do about it in future steps of the raw development pipeline. You can mask it behind a ton of sharpening, but this won't really fix it.
Photo of Todd Shaner

Todd Shaner, Champion

  • 1711 Posts
  • 580 Reply Likes
If you're trying to convince Adobe they need to improve LR and ACR demosaicing algorithms you'll need to provide fully edited image file comparisons to make your case. This includes applying equivalent Tone, Sharpening, and Noise Reduction controls, exporting to full-size TIFF file format, and providing share links to download and compare them. You'll also need to provide the raw file used for creating the TIFFs. If what you are claiming is correct the differences should be visible when viewed at 100% and obvious at 200% view.

Just a suggestion–Don't shoot the messenger!