AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

AmbiRefer3D: 3D Visual Grounding with Referential Ambiguity

Rongjiang Zhu, Wei Kang, Zeqi Liu, Chen junyu, Shuo Yang, Xinxiao Wu

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Visual grounding transformers

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add interactive disambiguation on top of D-LISAD-LISA · AmbiRefer3D real scenes (Mixture)Accuracy: 9.81 -> 41.57Improved
add interactive disambiguation on top of M3DRef-CLIPM3DRef-CLIP · AmbiRefer3D real scenes (Attribute + Spatial Relation)Accuracy: 33.24 -> 37.21Improved
add Lstop (stopping loss) (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: accuracy 42.36 -> 41.57, AvgR 3.74 -> 3.39Mixed
add Lcand (candidate ranking loss) (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: 38.13 -> 41.57Improved
add Lsel (constraint selection loss) (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: 24.47 -> 41.57Improved
use highest-scoring candidate as vtarget instead of random (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: 28.92 -> 41.57Improved
use highest-scoring candidate as vtarget instead of mean aggregation (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: 21.09 -> 41.57Improved
learned adaptive constraint selector instead of random (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: 23.31 -> 41.57Improved
learned adaptive selector instead of heuristic greedy (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureAccuracy: 33.76 -> 41.57Improved
apply top-3 constraints at once instead of one per round (ablation)interactive framework (D-LISA) · AmbiRefer3D MixtureS@1 / Accuracy: S@1 26.07 -> 30.42; Acc 41.57 -> 37.88Mixed
increase maximum interaction rounds (ablation)D-LISA + Ours · AmbiRefer3DAccuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy