Artefacts after verdandi

[Alex Fliker]

OS: Kubuntu 16.04 x86_64
Verdandi version: 2016.2.0

With --seam=hard I get this:
http://i.imgur.com/ETA3rPo.jpg

With --seam=blend I get this:
http://i.imgur.com/nLG6Z7t.jpg

Hard looks good if not considering seam lines. But what happened with Blend? Is there any clue?
If more information is needed just ask!

[tmodes (tmodes)]

This code is new. So there can be some bugs which occur only in special circumstances.
Can you provide the image files and the pto file to I can have a look, what goes wrong?

[Alex Fliker (fliker09)]

I uploaded files created by nona:
https://files.fm/u/4dxx8e9b

[tmodes (tmodes)]

First thanks. But such big files are bad for debugging. Can you reproduce the issue also with much smaller output images?
Also the pto file would be nice. For verdandi the blending order matters, the blending order can be calculated from the pto file.

[Alex Fliker (fliker09)]

I will try to reproduce the bug with down-scaled images. Regarding PTO... I don't work in GUI, only in terminal, so there is no sense in giving the PTO. Last actions in my pipeline is like this:
nona -m TIFF_m -z NONE --ignore-exposure -o panorama panorama.pto
verdandi --output=panorama.tif --wrap --seam=blend panorama0*.tif

[Alex Fliker (fliker09)]

I tried with a lot of different resolutions... Very inconsistent results. The image size has quite an effect on final result. One trend can be clearly observed - with lower resolution, seam line quality drops down. In the end I found working resolution (but on of the seam lines is not that good), but just by pure luck and random. Take a look at the links below:
https://filebin.net/e2wekguzhasuabwe/8192x4096.jpg
https://filebin.net/e2wekguzhasuabwe/12288x6144.jpg
https://filebin.net/e2wekguzhasuabwe/12728x6364.jpg

[tmodes (tmodes)]

Thanks for testing. But this does not help any further. Debugging with such big files does not work effectively.

PS: The 2 lines
nona -m TIFF_m -z NONE --ignore-exposure -o panorama panorama.pto
verdandi --output=panorama.tif --wrap --seam=blend panorama0*.tif
can be wrapped into one call to nona. As a side effect the big temporary files are not created (and beside the space you save the time to save and load these big files).

[Alex Fliker (fliker09)]

Ok, files are big, but the thing is - only they can reproduce the bug.

How to make the magic call to nona you mentioned?

[Alex Fliker (fliker09)]

An update on the bug - there is a dependency on geometry. I found this out while working on another sphere. Adjusting the control points in it helped me to fight off the artifact. So, it seems there is something to do with image features...

[tmodes (tmodes)]

I fixed some issues in the blending code in default branch. This should also fix the issue reported here.