TʀᴀᴄᴋGʀᴀᴘʜ constructs online open-vocabulary 3D scene graphs by propagating and tracking mask identities in image space using dense DINOv3-features before fusing them into persistent, class-agnostic 3D segments. 3D association reconciles tracking interruptions and revisits, while compact multi-view CLIP galleries enable language- and image-guided object search.
This repository contains the TʀᴀᴄᴋGʀᴀᴘʜ mapping core, which extends Hydra with online open-vocabulary mapping.
Follow the TʀᴀᴄᴋGʀᴀᴘʜ setup guide for Ubuntu 24.04 and ROS 2 Jazzy, including dataset and robot launch instructions. The guide imports all required repositories and selects the matching branches.
If you use this work in your research, please cite:
@article{hellesylt2026trackgraph,
title={{TRACKGRAPH}: Online Open-Vocabulary 3D Scene Graphs via Image-Space Tracking},
author={Hellesylt, Peder Borge and Gassol Puigjaner, Albert and Alexis, Kostas and Stahl, Annette},
journal={arXiv preprint arXiv:2609.31005},
year={2026},
url={https://arxiv.org/abs/2609.31005}
}Released under BSD-3-Clause.
This work was supported in part by:
- Research Council of Norway through NCEI (Grant No. 357451).
- European Commission through SYNERGISE under the Horizon Europe Programme (Grant No. 101121321).