Author Topic: Dense cloud results using 1.6.6 are better for my application than 1.8  (Read 393 times)


  • Newbie
  • *
  • Posts: 3
    • View Profile

I am working with stationary, terrestrial time-lapse photos where my intention is to capture real change using an approach similar to MEMI (Blanch et al., 2021). If you don't want to find the paper, I essentially align photos from multiple dates (between which some locations in the scene have changed), and then generate dense clouds using the tie points resulting from the combined alignment, but only the photos from one date at a time. I then compare the dense clouds to detect change between dates.

I would prefer to only use v1.8 as separate camera calibration folders are automatically created, which is really useful. Unfortunately, the dense cloud construction step in 1.8 leads to poor quality (nearly empty in some cases) dense clouds. I feel that this might be related to the updated depth map algorithm that was implemented in v1.7. That being said, I can still use v1.6 and get a dense cloud result that suits my needs (real change is effectively captured).

Does anyone have any insight into why this would be happening? If it is due to the new algorithm not being as tolerant of scene changes between photos, could a feature be implemented to allow the use of the older algorithms when desired? My concern is that if older versions become unusable at some point, my workflow will no longer be achievable.