HDRMerge: noise test

This is a noise test for HDRMerge. I used the exposure series from [Play Raw] Warning: Baroque can harm your eyes - #30 by vivo. This is how I came up with the sample images.

1. I opened the raws in HDRMerge with all options unchecked and exported them using 32/none. I forgot to export the masks; if necessary, I will do that later. Exposure bracketing is as follows:

name ev
1600 -1.5
1599 -1
1598 -0.5
1594 0
1595 0.5
1596 1
1597 1.5

Heeding @Morgan_Hardwood’s suggestion, I saved the following HDR files:

name raws
hdr2.dng 1600;1599
hdr3.dng 1600;1599;1598
hdr4.dng 1600;1599;1598;1595
hdr5.dng 1600;1599;1598;1595;1596

I omitted 1594 because it is 2 pixels off vertically, and developed 1597 by itself (I forgot to make an HDR file with 1600;1599;1598;1595;1596;1597).

2. Then I developed the 1597 and HDR files using PhotoFlow and cropped them using G’MIC for posting. The TIFs’ negative values are clipped but the +ve values are unbounded.

HDRMerge-noise.zip (8.7 MB)

Besides the unexpected noise levels of the HDR files, they should be aligned with 1597. Edit: oh, I see, the dimensions are different: 4000x3000 v 4036x3026.

Update Here are the hdr masks with the hdr6_.tif crop: hdr_masks_hdr6_.zip (8.3 MB).

name raws
hdr2.dng 1600;1599
hdr3.dng 1600;1599;1598
hdr4.dng 1600;1599;1598;1595
hdr5.dng 1600;1599;1598;1595;1596
hdr6.dng 1600;1599;1598;1595;1596;1597
1 Like

@afre I don’t see your point.

1594 (0 EV):

hdrmerge 1594-1600:

1 Like

Maybe I am doing it wrong, which is why I started this thread… I usually go from PhotoFlow to gmic. Since @gadolf is using dt, from dt with everything turned off:



hdr6.dng + 1.5EV

CRW_1597.DNG (default)

hdr6.dng (default) + 1.5EV

1 Like

@afre I don’t know what to say except that the only two settings I changed were checking align images at the beginning, and 10 for mask blur radius when saving the hdr, but I can’t see how this could relate to noise.
I just came from repeating it and I got the exact same results as my post above show.

Other thing is that I’m using this binary


from here

Isn’t that unusually high?

Maybe, but on my first set of pictures here in the forum that was the value that prevented artifacts in the highlights (clouds):

… and here

Since then, I’ve been using it as default radius

10 pixels:

3 pixels:

Conclusion: there’s no silver bullet. In this set of images, 3 pixels seems to be more suitable (maybe because there is more detail, maybe because there are no moving objects like clouds…). In these jpgs this is not very clear, but when editing the dngs in Darktable it’s also noticeable that the bigger radius messes with the boundary between darks and highlights
Good point!

1 Like

Thanks for doing this test. There is a bug in HDRMerge. We’ll fix it soon.


Here’s a dng created by a fixed hdrmerge from all 7 raw files using default settings.

I confess that I didn’t quite understood the original issue (since it seemed to work, to me), but anyway, glad to see hdrmerge is better, thanks! I am and will continue be a heavy hdrmerge user.

Where do I find this new version?

The fix is not yet committed. I’ll post here, as soon as it’s committed.

The build from 2018-04-05 you use, does not have this bug.

And this version 2018-04-05 isn’t the latest, correct? Will the latest (the one you’ve just built?) be in jcelaya’s repo?

If you click on the link you posted you will see that the current latest version is from 2018-05-14 (that’s the one with the bug).

The new latest version will be in the repo as soon as the fix is in repo. Currently the fix is on my machine. Will take a day or two.

1 Like

More precisely, the appimage will be available at most 24 hours after the commit, as it gets built automatically via a cron job that runs daily…

1 Like

I will wait for the Windows version. :wink:

1 Like

I just commited the fix. Should be in nightly builds soon.


uploaded at


The AppImage is ready: https://github.com/jcelaya/hdrmerge/releases/download/nightly/hdrmerge-git-20180609.glibc2.14-x86_64.AppImage


BTW, when should one use custom white level?