VOCA: Visual Odometry with Codec Awareness

ECCV 2026

Project page is under construction. Code, ablations, and results will be released upon paper publication.

Abstract

Camera pose estimation from image streams is a critical component of spatial world models that integrate perception for planning and decision making. Nearly all visual odometry (VO) and SLAM systems have focused on datasets containing raw and uncompressed videos. Many working systems, instead, use ubiquitous hardware units to compress and decode video streams efficiently, saving orders of magnitude in space and bandwidth. However, this lossy compression introduces visual artifacts that hinder the performance of traditional tracking systems. In this work, we present VOCA, a causal stereo visual-odometry method that exploits codec information to improve tracking performance. We achieve state-of-the-art performance on causal VO for relative trajectory error, efficiency, and absolute trajectory error on compressed streams.

Coming Soon

• Paper (arXiv)
• Code
• Ablations
• Evaluation results

BibTeX

@inproceedings{voca2026,
  title     = {VOCA: Visual Odometry with Codec Awareness},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026},
}

Project Page

The project page is hosted at this repository's GitHub Pages site: tum-vision.github.io/voca.