Hello jazvec,
I'm trying to figure out, what step in the version 1.2 you find much slower or working worse.
On the screenshot with the chunk info the matching time in the version 1.2 is 1 m 30 sec slower than in the version 1.1.6, but the number of keypoints is 50000 vs 30000, and the matching process takes longer time, as there are almost twice more feature points detected to be compared/matched.
The dense cloud generation step (without depth maps estimation) is 15 seconds longer.
And you have confirmed that the dense cloud generation on the same alignment takes almost the same time like in the version 1.1.6.
Maybe you can send us the sample set of images that we can process with the same parameters in both versions and see the huge difference in the processing time, that you are speaking about?
Note that High/Medium/Low accuracy for alignment in the new version has absolutely the same meaning as in the version 1.1.6. Highest option upscales the original images by two times by each side.