# SGAligner : 3D Scene Alignment with Scene Graphs

Source: [SGAligner : 3D Scene Alignment with Scene Graphs](<https://www.sota2.com/research/paper/68fa70ae1e209131cea4c7ad>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T04:12:40.329Z

Authors: Sayan Deb Sarkar, Ondrej Miksik, Marc Pollefeys, Daniel Barath, Iro Armeni

Published: 2023-04-28T14:39:22.000Z

## Abstract

Building 3D scene graphs has recently emerged as a topic in scene representation for several embodied AI applications to represent the world in a structured and rich manner. With their increased use in solving downstream tasks (eg, navigation and room rearrangement), can we leverage and recycle them for creating 3D maps of environments, a pivotal step in agent operation? We focus on the fundamental problem of aligning pairs of 3D scene graphs whose overlap can range from zero to partial and can contain arbitrary changes. We propose SGAligner, the first method for aligning pairs of 3D scene graphs that is robust to in-the-wild scenarios (ie, unknown overlap -- if any -- and changes in the environment). We get inspired by multi-modality knowledge graphs and use contrastive learning to learn a joint, multi-modal embedding space. We evaluate on the 3RScan dataset and further showcase that our method can be used for estimating the transformation between pairs of 3D scenes. Since benchmarks for these tasks are missing, we create them on this dataset. The code, benchmark, and trained models are available on the project website.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Scene Retrieval on 3RScan (test)](<https://www.sota2.com/research/sota/scene-retrieval-on-3rscan-test.md>): Scene Retrieval; 3RScan (test); MRR: 95.

- [3D Scene Graph Alignment on 3RScan](<https://www.sota2.com/research/sota/3d-scene-graph-alignment-on-3rscan.md>): 3D Scene Graph Alignment; 3RScan; R@top-2: 96.4.

- [Point Cloud Mosaicking on 3RScan (143 scenes)](<https://www.sota2.com/research/sota/point-cloud-mosaicking-on-3rscan-143-scenes.md>): Point cloud mosaicking; 3RScan (143 scenes); Accuracy: 0.94.

- [3D Point Cloud Mosaicking on 3RScan](<https://www.sota2.com/research/sota/3d-point-cloud-mosaicking-on-3rscan.md>): 3D Point Cloud Mosaicking; 3RScan; Accuracy: 1.215.

- [Scene Graph Alignment on 3RScan Scenario i (sub-scene on the original scan (no changes))](<https://www.sota2.com/research/sota/scene-graph-alignment-on-3rscan-scenario-i-sub-scene-on-the-original-scan-no-changes.md>): Scene Graph Alignment; 3RScan Scenario i (sub-scene on the original scan (no changes)); MRR: 97.9.

- [Scene Graph Alignment on 3RScan Scenario ii 3D sub-scene on a rescan (with changes)](<https://www.sota2.com/research/sota/scene-graph-alignment-on-3rscan-scenario-ii-3d-sub-scene-on-a-rescan-with-changes.md>): Scene Graph Alignment; 3RScan Scenario ii 3D sub-scene on a rescan (with changes); MRR: 93.6.

- [Scene Graph Alignment on 3RScan Scenario iii (sub-scenes containing changes)](<https://www.sota2.com/research/sota/scene-graph-alignment-on-3rscan-scenario-iii-sub-scenes-containing-changes.md>): Scene Graph Alignment; 3RScan Scenario iii (sub-scenes containing changes); MRR: 88.8.

- [3D Point Cloud Registration on 3RScan (test)](<https://www.sota2.com/research/sota/3d-point-cloud-registration-on-3rscan-test.md>): 3D Point Cloud Registration; 3RScan (test); CD: 0.0111.

- [Overlap Check on 3RScan (val)](<https://www.sota2.com/research/sota/overlap-check-on-3rscan-val.md>): Overlap Check; 3RScan (val); Precision: 92.03.

- [Scene Graph Node Alignment on 3RScan (val)](<https://www.sota2.com/research/sota/scene-graph-node-alignment-on-3rscan-val.md>): Scene Graph Node Alignment; 3RScan (val); Mean RR: 96.3.

- [Overlap Check on 3RScan](<https://www.sota2.com/research/sota/overlap-check-on-3rscan.md>): Overlap Check; 3RScan; Precision: 93.29.

- [3D Scene Graph Generation on MA3DSG-Bench SCP setting 1.0 (test)](<https://www.sota2.com/research/sota/3d-scene-graph-generation-on-ma3dsg-bench-scp-setting-1-0-test.md>): 3D scene graph generation; MA3DSG-Bench SCP setting 1.0 (test); Triplet Recall@1: 22.5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2304.14880>)

[PDF](<https://arxiv.org/pdf/2304.14880.pdf>)

[Implementation](<https://www.sota2.com/sgaligner.github.io>)
