@inproceedings{li-etal-2026-origround,
title = "{O}ri{G}round: Orientation-Aware Neuro-Symbolic Zero-Shot 3{D} Visual Grounding",
author = "Li, Haochen and
Shi, Jiaxin and
Liu, Ruonan and
Liufu, Luo",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2237/",
pages = "47877--47897",
ISBN = "979-8-89176-462-0",
abstract = "Zero-shot 3D visual grounding (3DVG) aims to localize a referred object in a 3D scene without task-specific grounding supervision. Existing training-free methods commonly reason over object categories, centers, and bounding boxes, but these representations do not encode the local reference frames needed to interpret viewpoint-dependent expressions such as left of, behind, or facing the door. We propose OriGround, an orientation-aware neuro-symbolic framework that explicitly models object-centric orientations and infers observer viewpoints for zero-shot 3DVG. OriGround parses each query into a viewpoint-aware symbolic program, estimates object-centric orientation bases from multi-view observations, performs anchor- and observer-specific geometric relation scoring, and constructs a reference-frame-aligned visual prompt for final VLM-based disambiguation. On Nr3D, OriGround improves over the strongest training-free baseline by 8.4 points in overall accuracy, with larger gains of 8.9 points on hard queries and 5.6 points on view-dependent queries. On ScanRefer, OriGround further improves overall Acc@0.5 by 17.3 points over VLM-Grounder. Project page: https://origround.github.io/"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-origround">
<titleInfo>
<title>OriGround: Orientation-Aware Neuro-Symbolic Zero-Shot 3D Visual Grounding</title>
</titleInfo>
<name type="personal">
<namePart type="given">Haochen</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxin</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruonan</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Luo</namePart>
<namePart type="family">Liufu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Zero-shot 3D visual grounding (3DVG) aims to localize a referred object in a 3D scene without task-specific grounding supervision. Existing training-free methods commonly reason over object categories, centers, and bounding boxes, but these representations do not encode the local reference frames needed to interpret viewpoint-dependent expressions such as left of, behind, or facing the door. We propose OriGround, an orientation-aware neuro-symbolic framework that explicitly models object-centric orientations and infers observer viewpoints for zero-shot 3DVG. OriGround parses each query into a viewpoint-aware symbolic program, estimates object-centric orientation bases from multi-view observations, performs anchor- and observer-specific geometric relation scoring, and constructs a reference-frame-aligned visual prompt for final VLM-based disambiguation. On Nr3D, OriGround improves over the strongest training-free baseline by 8.4 points in overall accuracy, with larger gains of 8.9 points on hard queries and 5.6 points on view-dependent queries. On ScanRefer, OriGround further improves overall Acc@0.5 by 17.3 points over VLM-Grounder. Project page: https://origround.github.io/</abstract>
<identifier type="citekey">li-etal-2026-origround</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2237/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>47877</start>
<end>47897</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T OriGround: Orientation-Aware Neuro-Symbolic Zero-Shot 3D Visual Grounding
%A Li, Haochen
%A Shi, Jiaxin
%A Liu, Ruonan
%A Liufu, Luo
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F li-etal-2026-origround
%X Zero-shot 3D visual grounding (3DVG) aims to localize a referred object in a 3D scene without task-specific grounding supervision. Existing training-free methods commonly reason over object categories, centers, and bounding boxes, but these representations do not encode the local reference frames needed to interpret viewpoint-dependent expressions such as left of, behind, or facing the door. We propose OriGround, an orientation-aware neuro-symbolic framework that explicitly models object-centric orientations and infers observer viewpoints for zero-shot 3DVG. OriGround parses each query into a viewpoint-aware symbolic program, estimates object-centric orientation bases from multi-view observations, performs anchor- and observer-specific geometric relation scoring, and constructs a reference-frame-aligned visual prompt for final VLM-based disambiguation. On Nr3D, OriGround improves over the strongest training-free baseline by 8.4 points in overall accuracy, with larger gains of 8.9 points on hard queries and 5.6 points on view-dependent queries. On ScanRefer, OriGround further improves overall Acc@0.5 by 17.3 points over VLM-Grounder. Project page: https://origround.github.io/
%U https://aclanthology.org/2026.findings-emnlp.2237/
%P 47877-47897
Markdown (Informal)
[OriGround: Orientation-Aware Neuro-Symbolic Zero-Shot 3D Visual Grounding](https://aclanthology.org/2026.findings-emnlp.2237/) (Li et al., Findings 2026)
ACL