Bug #10662

Exported Image not equally as sharp as what is seen in darkroom mode

Added by Joe Giampaoli over 3 years ago. Updated over 3 years ago.

Closed: invalid
Start date:
Due date:
% Done:


Affected Version:
hardware architecture:


I have been seeing this for a while already. For some time I thought that it was probably my imagination, but today as I just exported an image I noticed that the final image lacks a certain "sharpness" or "detail" as what I see while editing the image in darkroom mode. This lack of detail can be seen in both the image preview thumbnail in lighttable (if seen at largest possible, single image thumbnail) and when seeing the exported image in image viewer or even in GIMP.

If you see the thumbnail Darkroom_Mode.png you will clearly see there is more hair details in the spider, these detail is absolutely not as good in the other screenshots I made.

Something tells me it's more of a darkroom thing...

Sending you screenshots of how it looks in different ways.

This happens with both RAW and JPG files.

I'm using Debian Jessie 64bit

Any clues?


Joe Giampaoli

Exported_Image.png - How I see it when exported in image viewer (478 KB) Joe Giampaoli, 10/12/2015 02:27 AM

Darkroom_Mode.png - How I see it in Darkroom mode (655 KB) Joe Giampaoli, 10/12/2015 02:27 AM

Lighttable_Mode.png - How I see it in Lighttable mode (439 KB) Joe Giampaoli, 10/12/2015 02:27 AM

GIMP.png - How I see the exported image in GIMP (676 KB) Joe Giampaoli, 10/12/2015 02:27 AM

core_options.png (46.7 KB) Joe Giampaoli, 10/12/2015 03:26 AM

darkroom_mode.png (880 KB) Joe Giampaoli, 10/12/2015 03:26 AM

exported.png (899 KB) Joe Giampaoli, 10/12/2015 03:26 AM

lighttable_mode.png (685 KB) Joe Giampaoli, 10/12/2015 03:26 AM


#1 Updated by Joe Giampaoli over 3 years ago

My wrong, I think this only happens with RAW's not sure, I've been testing bit more with other images...

#2 Updated by Joe Giampaoli over 3 years ago

OK, I just changed settings in "Core Options" to use bilinear in both:

1 Demosaicing for zoomed out darkroom mode
2 Pixel Interpolator

Seems to have better results, at least I can now see in darkroom mode exactly what will be exported, should these settings be correct?

Attaching other screenshots, you can see that they are more alike.

#3 Updated by Tobias Ellinghaus over 3 years ago

Did you try using lanczos3? For me that gives good results. And for darkroom I set it to "full". It seems to be fast enough.

#4 Updated by Pascal de Bruijn over 3 years ago

  • Status changed from New to Closed: invalid

Unless I'm missing something, these comparisons are quite irrelevant.

Since you're comparing darktable's downsampling vs the downsampling of your favorite image viewer (whatever that might be).

darktable defaults to a relatively high quality (which produces fairly sharp results) lanczos algorithms.

Many image image viewers use speed optimized algorithms like bilinear (which tends to be a tad blurry), or nearest-neighbor (which is often sharp, but might be perceived as noisy).

Regardless, any downsampling algorithm will affect a number of interrelated image qualities like moire, detail/sharpness and noise, which are traded off against each other and speed. Practical effects of this vary depending on the source imagery, as many issues only surface with sharp strongly textured source imagery.

Considering you have no control over which image viewer (keep in mind a browser is just as well a potentially downsampling image viewer) your client might be using, it seems incredibly unwise to adapt your workflow to any specific settings in an attempt to "match this up".

If you're insistant on doing any sharpness evaluation, the only way to do this (semi-)meaningfully is to compare at 1:1 pixel zoom.

@Darkroom/Lighttable mode, keep in mind that lighttable is speed optimized, and uses our own internal thumbnails, for detailed evaluations of any kind you should always use Darkroom mode.

Also, darktable has a HQ export option, that keeps the imaging pipeline at full resolution until the very last minute, which in some cases might preserve a tad more detail, but it will slow exports down considerably (which is why it's disabled by default).

#5 Updated by Joe Giampaoli over 3 years ago


You are correct in the way that the image viewer was probably using another algorithm! That was my first impression too. unfortunately it's not, that is why I also saw these same results in GIMP which is not your common image viewer and for which I also included the screenshots for. I saw the same results in it! So it's clearly not an image viewer problem and their way of representing the pixels. I also compared with imagemagick viewer too, for that I did not include screenshots. But it was there!

The whole thing about this is that it's not a 1:1 pixel comparison, actually I try not to be much of a "pixel peeper", noise and grain are even sometimes things I can live with to some extent. The point here is that if I'm editing a photo and happy with the results I see in DT and then see something different in my final exported image, well that's not good! I could even see the difference in downscaled exported images, they were just not as sharp as what I was seeing during the editing process....

Anyway, I seem to have fixed it by applying the settings I did, now my exported images look a bit "truer" to what I was working with.

#6 Updated by Joe Giampaoli over 3 years ago

Tobias Ellinghaus wrote:

Did you try using lanczos3? For me that gives good results. And for darkroom I set it to "full". It seems to be fast enough.

I did, and I found out that lanczoz3 should be used when applying a bit of rotating, otherwise the image would get that "blurry" look when exported and even during the editing. So for me lanczos3 in pixel interpolation and always bilinear (fast) for demosaicing seem to give me the truest results.


Also available in: Atom PDF