Problem in Reconstrucition with Roma and Loftr

[Asnly1]

Hi. Thank you for your fantastic work.

I have trouble with reconstrution by using Roma and Loftr. Currently, I am trying to reconstruction a scene with a book in a round table. When using original colmaps, it can utilize all 63 jpg photos and reconstruct 63 camera poses. Moreover, when I try to use sift+kornia_mathcer, it can also utilize all 63 photos and reconstruct 63 camera poses (INFO:deep-image-matching:Largest model is #0 with 63 images.) . However, when I use Roma or Loftr, it can only utilize 2 photos and reconstruct 2 camera poses (INFO:deep-image-matching:Largest model is #0 with 2 images.) . I list my settings below:

63 Photos. Colmap; Strategy: exhaustive; Result: 63 photos and 63 camera poses Sift+kornia_mathcer; Strategy: sequential; Quality: high; Result: 63 photos and 63 camera poses Roma; Strategy: sequential; Quality: high; Result: 2 photos and 2 camera poses Roma; Strategy: sequential; Quality: medium; Result: 2 photos and 2 camera poses Roma; Strategy: sequential; Quality: low; Result: 2 photos and 2 camera poses Roma; Strategy: bruteforce; Quality: high; Result: 2 photos and 2 camera poses Roma; Strategy: bruteforce; Quality: medium; Result: 2 photos and 2 camera poses Roma; Strategy: bruteforce; Quality: low; Result: 2 photos and 2 camera poses Loftr; Strategy: sequential; Quality: high; Result: 2 photos and 2 camera poses Loftr; Strategy: sequential; Quality: medium; Result: 2 photos and 2 camera poses Loftr; Strategy: sequential; Quality: low; Result: 2 photos and 2 camera poses

Is there something wrong with my settings?

I have attached some of my my reconstruction scene. I simply walk around the table.

[lcmrl]

Hi, thanks for reporting. Which branch are you using? If not try with the 'dev' branch. If you still have the problem let me know, momently using the branch 'SRIF' will fix your problem

[lcmrl]

I close the issue for inactivity, please feel free to reopen if you still encounter the problem