Open LGimmick opened 2 years ago
Hmm, I also received other report of similar problem.
Can you send me a test image set that reproduces the problem? If it is too large for upload here, you can mail me at petteri.aimonen (at) devembedded.com
I've sent you some images :).
@LGimmick Thanks for the logs. The key line was "Valid area not defined for aligned_Object3.bmp, using default.".
The OpenCL-accelerated code was not setting the valid area, so the bug only appeared when OpenCL was available. Fixed in commit 881e6b53bd62.
There is new build for testing here: https://github.com/PetteriAimonen/focus-stack/releases/tag/latest
New version works for me, too.
Thank you! Works perfectly fine now.
Hi,
I have tried your program and I think the result is very good and it's realy fast. But I noticed something concerning the crop-function:
The crop only works if the image content moves to the right - so the offset is oriented left. Otherwise the mirrored image content on the left side stays in the final image. [Global-Alignment and changed reference image does fix this, not sure about vertical movement]. And I have a few image-sets without any content-movement. The input-images have a resolution of 1600x1200 px. No matter what parameters I use, all aligned images (even the first one with reference = 0 and global-align) are created with 1664x1216 px and the result image contains the mirrored content on all edges, too. I guess it's kind of the same effect and problem seems to happen somewhere in the calculation of the area with full data.