Skip to content

Conversation

@sarlinpe
Copy link
Collaborator

@sarlinpe sarlinpe commented Mar 3, 2024

  • x100 faster pose estimation by optimizing recover_matches_2d3d
  • feature matching with LightGlue
  • Absolute pose estimation for single-image and rig with PoseLib - slightly better than pycolmap
LIN
  query_phone
    pycolmap: 57.1 / 71.5
    poselib:  60.0 / 72.0
  query_hololens
    pycolmap: 69.2 / 83.4
    poselib:  70.5 / 82.7
HGE
  query_phone
    pycolmap: 45.5 / 60.8
    poselib:  46.7 / 60.6
  query_hololens
    pycolmap: 56.6 / 69.4
    poselib:  57.4 / 69.5

cc @vlarsson


import pycolmap
try:
import poselib
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe adding the installation instructions in README, perhaps also updating the pyproject and Dockerfile. Same for LightGlue.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think LightGlue comes with hloc 1.4 already (?)

@sarlinpe
Copy link
Collaborator Author

Poke @mihaidusmanu

# avoid duplicate observations
if p3d_id != -1 and p3d_id in p2d_to_p3d[i]:
for match_idx, ((i, j), p3d_id) in enumerate(zip(matches, p3d_ids)):
if p3d_id != -1 and (i, p3d_id) in p2d_to_p3d:
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit, I guess this is guaranteed to be != -1? So would maybe assert it?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(and maybe the check should be against invalid_point3d_id?)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants