You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I want to give an update on what future changes we are planning for COLMAP. We want to make it:
More research-friendly, i.e. more modular, easier to extend and to hack
More robust, by leveraging recent research, for example, on learning-based matching, self-calibrating minimal solvers, etc.
Faster, with more parallelization when possible
Cover a broader set of use cases, including multi-sensor reconstruction (multi-camera rigs, spherical cameras, GPS, depth), long videos, etc.
We think that progressively moving the higher-level logic to Python will make these easier to achieve, while keeping a performant core C++ library. To get there, we've derived a first set of smaller milestones:
Would benefit from benchmarking solvers that estimate the camera calibration
Two-view verification: support the least-squares refinement of E/H/F matrices via Ceres
Create and benchmark a database format based on HDF5 as alternative (and later replacement) to SQL
Design and implement the serialization of reconstructions via protocol buffer, as alternative (and later replacement) to the custom bin/text format
Switch some internal data types (images, keypoints) to plain Eigen matrices for better interoperability with Python
More extensive coverage of the Python bindings: database, bitmap, estimators
Automated documentation generation for the Python bindings
Proper & tested integration of multi-camera rigs into the pipeline (to later support spherical images via tiling)
Proper & tested integration of GPS into the pipeline (maybe a rework of Sfm-GPS BA #1409)
It would be great to get feedback from the community (especially power users) on whether these goals align with their use cases. If some folks have the bandwidth to help on some of these points, you are very much welcome to contribute. We will then be able to tackle deeper changes.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hello everyone,
I want to give an update on what future changes we are planning for COLMAP. We want to make it:
More research-friendly, i.e. more modular, easier to extend and to hack
More robust, by leveraging recent research, for example, on learning-based matching, self-calibrating minimal solvers, etc.
Faster, with more parallelization when possible
Cover a broader set of use cases, including multi-sensor reconstruction (multi-camera rigs, spherical cameras, GPS, depth), long videos, etc.
We think that progressively moving the higher-level logic to Python will make these easier to achieve, while keeping a performant core C++ library. To get there, we've derived a first set of smaller milestones:
It would be great to get feedback from the community (especially power users) on whether these goals align with their use cases. If some folks have the bandwidth to help on some of these points, you are very much welcome to contribute. We will then be able to tackle deeper changes.
Thanks!
cc @ahojnnes @tsattler @vlarsson @mihaidusmanu @Phil26AT @arjunkarpur @B1ueber2y @Zador-Pataki
Beta Was this translation helpful? Give feedback.
All reactions