Reproduce “A Photometrically Calibrated Benchmark For Monocular Visual Odometry” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jakob Engel, Vladyslav Usenko, Daniel Cremers

The TUM monoVO dataset introduces a new benchmark for monocular visual odometry and SLAM, featuring detailed photometric calibration information for over 100 minutes of diverse real-world video and a scalable loop-closure-based evaluation. It provides resources for developing photometric-aware algorithms and demonstrates that Direct Sparse Odometry (DSO) generally achieves higher accuracy and robustness than ORB-SLAM across varying conditions.

Key insights

  • Accurate non-parametric photometric calibration for camera response and vignetting can be achieved using large data volumes, avoiding explicit smoothness priors or uniform surfaces.
  • A loop-closure based drift evaluation, using the `e_align` metric, offers a robust and scalable method for assessing monocular VO/SLAM accuracy on large datasets without full external ground truth.
  • Algorithm performance in monocular VO/SLAM is strongly influenced by factors like camera field of view and resolution, with direct methods exhibiting better robustness to resolution changes compared to feature-based methods.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.