I'm impressed. Especially for a completely free solution.FWIW, darktable with just filmic, exposure, color balance, color calibration, and tone equalizer. Export fed through nind-denoise, then GMIC's RL-deblur.
I see plenty of artifacts, but nind-denoise is still in its early stage, so I'm not complaining
Yes I fully agree with you. Far less detail here.i did you a second version.
this is converting raf to dng with iridient transformer (set to smoother raw process, and minimal chroma end luma noise reduction)
still imported in LR, noise reduction and sharpen added in LR (works much better than working on original raf in LR, iridient conversion has a great demosaicing algorithm)
again basic correction, and subject masking to isolate background and add more noise reduction to background.
basically same process as before but starting with iridient conversion
spoiler, deeprime work much better
Agreed, the "profiled denoise" in darktable is just generic, while it's not bad, it cannot compete with the "smart" AI/neural-network tools. That's why the nind-denoise's author is looking into having the tool integrated into darktable workflow, too.What was still missing, a darktable take.
12800 ISO, darktable, denoise profiled + raw denoise
IMO it cannot compete against professional denoising tools. Maybe others can get better result with darktable.
The detail looks good here, but I don't really see a lot of noise reduction going on.Agreed, the "profiled denoise" in darktable is just generic, while it's not bad, it cannot compete with the "smart" AI/neural-network tools. That's why the nind-denoise's author is looking into having the tool integrated into darktable workflow, too.What was still missing, a darktable take.
12800 ISO, darktable, denoise profiled + raw denoise
IMO it cannot compete against professional denoising tools. Maybe others can get better result with darktable.
For high-ISO shots, I prefer leaving the fine grains in. They not only help with retaining details, they keep the perceived sharpness, as well as simulating the old film grain (addressing only chroma noise, not luma noise). Also, when you resize the image to a smaller size, the grains will be gone while the detail and sharpness contribute to the sharpness of the smaller image.
darktable, profiled denoise in wavelets mode
heh, you know I prefer retaining details over clean image
Although the vanilla RAW doesn't have any clipping in the white fur, there isn't a lot of details in there. I used to compress the DR a lot, especially the latest filmic has the option "safe" in "contrast highlight" to protect the highlights from clipping. However, I found without those "clipped" highlights, the image looks dull with low contrast. Thus, for most of my recent images, I tend to allow clippings on unimportant spots, giving the image a bit more contrast globally.Also, just curious, can anything be done in Darkatable to recover any more detail in the white fur? Yours is significantly better than Martin's on that score, but it falls a bit short of what I managed with Lightroom. The green channel is good and blown, but the other two channels are intact and still retain a bit of detail to exploit. Lightroom, despite some other shortcomings, tends to be quite capable in that department.
Unfortunately, the histogram in your editor only represents the image as it's currently rendered, not what the RAW file actually looks like (same as mine) This is the True RAW histogram via RawDigger (Note the green channel clipping)...heh, you know I prefer retaining details over clean image
Although the vanilla RAW doesn't have any clipping in the white fur, there isn't a lot of details in there. I used to compress the DR a lot, especially the latest filmic has the option "safe" in "contrast highlight" to protect the highlights from clipping. However, I found without those "clipped" highlights, the image looks dull with low contrast. Thus, for most of my recent images, I tend to allow clippings on unimportant spots, giving the image a bit more contrast globally.Also, just curious, can anything be done in Darkatable to recover any more detail in the white fur? Yours is significantly better than Martin's on that score, but it falls a bit short of what I managed with Lightroom. The green channel is good and blown, but the other two channels are intact and still retain a bit of detail to exploit. Lightroom, despite some other shortcomings, tends to be quite capable in that department.
LR/IXT/Topaz DeNoise
I found the same with shadow. For theater shots, most of the details are in the dark, thus, I preserve the shadows in darktable to show them, but for a typical shot, I would go for a little punchier look. While I've mentioned that I don't like the film simulations as they clipped highlights and shadows, I have to admit that I'm learning quite a bit from them. While they are technically inferior to compressing DR, they give the image a more pleasant look.
There's also the "shadows and highlights" module that can boost the details/contrast in shadow and highlight, but I don't generally use it as it can create an unnatural look (to me).
I guess it all comes down to personal preference.
although the highlights aren't clipped in vanilla RAF, there isn't a lot of interesting details neither
without a clipped spot (serving as a white-reference point), the image looks a little bit duller to me even with the contrast bumped up a bit
In DXO the way is to edit the gradation:
Some detail is lost, but not especially good detail. Looks good.In DXO the way is to edit the gradation:
This should work in other applications too.
The dynamic range of the image is larger than the range of 8 bit JPEG, so some compression for the dynamic range is needed.
Result:
Yes.
Much appreciated.
You're right. Yours has more grain but looks better than mine.
Wow really cool what you get out of that 12800 ISO image! Astonishing.
Really a great result, as I already said when replying to your other post in this thread.
ah, that's where I got confused. The details are obviously clipped in there, yet the histogram still has some room on the right.Unfortunately, the histogram in your editor only represents the image as it's currently rendered, not what the RAW file actually looks like (same as mine) This is the True RAW histogram via RawDigger (Note the green channel clipping)...
and where it's clipping...
If the details weren't there to begin with, I guess it's up to the reconstruction algorithm to interpolate the data. I don't know darktable well enough to explore further, but perhaps an AI/neural-network is the way to go in terms of generating/constructing data.
There is is detail that is otherwise being lost in your screenshot. Can you you work on that area locally?
Thank you. As the OP, I did an intermediate assessment.
Very helpful - many thanks.
I have been following this thread avidly as I also prefer to use linux and would prefer not to have to dive back into windows if I can help it.
Yes like me
Interesting, so you have a different workflow. In my current workflow, export & denoising comes at the end.
We should not finally judge from one single image. But yes, from what I've seen here in this thread, I must admit that's about my own perception, too. I believe Pablo with his fine hair is pretty challenging at ISO 12800. As far as I see, the AI based tools do a better job with Pablo at ISO 12800 more or less from scratch as compared to Neat Image. This is also true for your result. Neat Image is still not bad, it is still much better than for example darktable alone, but I had to play for quite a while with it until I had a satisfactory Pablo that came somehow close to what the AI based tools deliver, and it was still not the same. OTOH, I have already successfully denoised a number or more "normal" images with Neat Image that I cannot publish, and I am happy with the results. Those were more in the ISO 4000-6400 region and not as challenging as Pablo. But if I had a Windows machine for my photo work, there is a good chance that I'd have preferred Topaz DeNoise AI over Neat Image. The price is about the same. Topaz even had a discount recently.
You're welcome. Many thanks for participating in the challenge.
