Slightly odd brightness changes in stitched image

I don’t know if this is just user error, or something I’m not aware of - but after posting this image: [Capture Challenge] People without people - #64 by 123sg
I noticed that while the stitching seems absolutely perfect, there seems to be some slightly odd brightness changes happening on the tree trunk and a background tree:

It almost looks like I applied some kind of tonemapping to the original files, but I didn’t.

This is the same image stitched in MICE, with a dodgy join, but the tones/brightness are as the original files. See how the trunk is much darker?

Any thoughts anyone?

It’s hard to beat MICE :slight_smile:

You can try debug mode CTRL + D and then in stitching options switch to the OpenCV blending algorithm. The details won’t be as good as with Multiblend though.

Also feel free to send me a few images where I could test this.

You can also thumbs up this related issue in OpenCV that I put up: MultiBandBlender loss of fine details · Issue #24073 · opencv/opencv · GitHub

1 Like

Thanks Tomas! I’ll try debug mode and the other algorithm later today.

Well, xpano’s stitching well and truly beat MICE in terms of positioning accuracy - nothing I could do in MICE could get rid of that unsightly misalignment on the roof. :smiley:

1 Like

I’ll try and sort out some images for you too.

I just tried that - sure enough solved my original issue, but it goes very fuzzy in the shadows. I think I prefer the original problem!

Here are the source images for the pano above - licenced with the usual Play Raw Creative Commons, By-Attribution, Share-Alike.
K10d pano (38.0 MB)
Hope they might be useful.
Finally, thanks for your work! Regardless of my issue above it’s great software.

1 Like

By the way, @krupkat, have you seen this project? 'stitching' - a foss python cli to generate panoramas from single images

I’m assuming there’s not really any overlap with xpano - but I guess you might find it interesting.

Thanks for the nice feedback! Really glad that the app is useful for you! I’m able to replicate your results with Xpano - some weird brightness changes with Multiblend and bad details with the OpenCV blener.

I saw the OpenStitching project before, yes. Now I have tested it as well, but didn’t manage to stitch the images you uploaded in a quick test with it.

Looking at its code, it is using the OpenCV MultiBand blender, so it would actually benefit from solving MultiBandBlender loss of fine details · Issue #24073 · opencv/opencv · GitHub as well, we may be able to join forces there @lukasalexanderweber, at least with upvoting the issue :slight_smile:

1 Like

I’ve got this result using stitch * --final_megapix 0.5 --adjuster reproj.

@123sg I think Xpano and the stitching package have lot’s of overlap, both heavily rely on opencvs stitching module. @krupkat I think Xpano is mainly a gui helping users by not having to use a command line. stitching is a python wrapper around the library focusing on being able to get also all intermediate results.

@krupkat I’ve upvoted your issue :slight_smile: