Research map / Point clouds / Point cloud classification
WildRefer: 3D Object Localization in Large-scale Dynamic Scenes with Multi-modal Visual Data and Natural Language
Zhenxiang Lin, Xidong Peng, Peishan Cong, Ge Zheng, Yujing Sun, Yuenan HOU, Xinge Zhu, Sibei Yang, Yuexin Ma
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Dynamic Visual Encoder (temporal attention) to the baseline (ablation) | WildRefer baseline · STRefer | Acc@0.5: 45.12 -> 50.28 | Improved |
| add Triple-modal Feature Interaction (image fusion) on top of DVE (ablation) | WildRefer · STRefer | Acc@0.5: 50.28 -> 54.97 | Improved |
| add DVE to baseline (ablation) | WildRefer baseline · LifeRefer | Acc@0.25: 32.46 -> 38.27 | Improved |
| add TFI on top of DVE (ablation) | WildRefer · LifeRefer | Acc@0.25: 38.27 -> 38.89 | Improved |
| fuse language with point features first, then images (vs Vision First) (ablation) | TFI · STRefer | Acc@0.5: 52.72 -> 54.97 | Improved |
| point-dominated rather than image-dominated fusion (ablation) | TFI · STRefer | Acc@0.25: 52.25 -> 62.01 | Improved |
| fusion order variants (ablation) | TFI · LifeRefer | Acc@0.25: Image Domin. 38.82 vs ours 38.82 | No clear effect |
| use 2 frames instead of 1 in DVE (ablation) | DVE · STRefer | Acc@0.25: 59.38 -> 62.01 | Improved |
| increase frames from 2 to 3 or 4 (ablation) | DVE · STRefer | Acc@0.25: 62.01 -> 59.94 (3 frames), 60.41 (4 frames) | Hurt |
| replace DVE with feature concatenation across frames (ablation) | DVE · STRefer | Acc@0.25: 62.01 -> 40.24 | Hurt |
| one-stage instead of two-stage pipeline for BUTD-DETR | BUTD-DETR · LifeRefer | Acc@0.25: 30.81 -> 32.46 | Improved |
| WildRefer vs best prior method on STRefer | WildRefer · STRefer | Acc@0.5: 47.47 -> 54.97 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- 3D-SPS — 3D-SPS: Single-Stage 3D Visual Grounding via Referred Point Progressive Selection (CVPR 2022) · better on 1 of 1 shared result
- 3DJCG — 3DJCG: A Unified Framework for Joint Dense Captioning and Visual Grounding on 3D Point Clouds (CVPR 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3EED: Ground Everything Everywhere in 3D (NeurIPS 2025 · similarity 0.87)
- G^3-LQ: Marrying Hyperbolic Alignment with Explicit Semantic-Geometric Modeling for 3D Visual Grounding (CVPR 2024 · similarity 0.87)
- From Objects to Anywhere: A Holistic Benchmark for Multi-level Visual Grounding in 3D Scenes (NeurIPS 2025 · similarity 0.87)
- CoT3DRef: Chain-of-Thoughts Data-Efficient 3D Visual Grounding (ICLR 2024 · similarity 0.87)
- AugRefer: Advancing 3D Visual Grounding via Cross-Modal Augmentation and Spatial Relation-based Referring (AAAI 2025 · similarity 0.87)
- DenseGrounding: Improving Dense Language-Vision Semantics for Ego-centric 3D Visual Grounding (ICLR 2025 · similarity 0.86)
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.
