Research map / Video / Video segmentation and tracking
MoT3DVG: A Benchmark for Outdoor 3D Visual Grounding with Motion-Aware Descriptions and Temporal Cues
Shijia Zhao, Xusheng Guo, Qiming Xia, Xun Huang, KeZheng Xiong, Zhihong Liu, Chenglu Wen
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces MoT3DVG, a benchmark for outdoor 3D visual grounding that uses motion-aware language descriptions and temporal cues to address static-object limitations of existing indoor 3DVG research.
Problem
Existing 3D visual grounding research focuses on static indoor objects, which limits its use in dynamic outdoor autonomous driving scenes where motion descriptions capture the temporal dynamics of moving objects.
Approach
A benchmark pairing outdoor 3D scenes with motion-aware referring descriptions and temporal cues for grounding moving objects; the abstract is truncated so the specific method details are not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SLIDE temporal encoding of historical point clouds | 3EED-style grounder · MoT3DVG overall | Acc@0.5m: 54.94 -> 57.34 | Improved |
| add SLIDE temporal encoding | 3EED-style grounder · MoT3DVG Unique subset 0-20m | Acc@1.0m: 57.08 -> 50.68 | Hurt |
| add SLIDE temporal encoding | 3EED-style grounder · MoT3DVG Multiple subset | Acc@0.5m: 47.96 -> 60.25 | Improved |
| add short-term window encoding only (ablation) | DynaVG · MoT3DVG overall | Acc@0.5m: 54.94 -> 55.20 | Improved |
| add long-term window shifting on top of STWE (ablation) | DynaVG · MoT3DVG overall | Acc@0.5m: 55.20 -> 57.34 | Improved |
| use SLIDE instead of concatenating historical frames (ablation) | 3EED · MoT3DVG overall | Acc@0.5m: 56.04 -> 57.34 | Improved |
| use fine-grained temporal motion descriptions instead of coarse motion (ablation) | DynaVG · MoT3DVG overall | Acc@0.5m: 54.76 -> 57.34 | Improved |
| remove motion attributes from prompts (ablation) | DynaVG · MoT3DVG overall | Acc@0.5m: 57.34 -> 54.33 | Hurt |
| remove appearance attributes from prompts (ablation) | DynaVG · MoT3DVG overall | Acc@0.5m: 57.34 -> 58.63 (Acc@1.0 69.66 -> 69.15) | No clear effect |
| add RGB image semantics to LiDAR input (ablation) | DynaVG · MoT3DVG overall | Acc@0.5m: 57.34 -> 58.20 | Improved |
| add adaptive temporal gating (ablation) | DynaVG · MoT3DVG Unique and long range | Acc/mIoU: 0-20m overall Acc@0.5 64.87 -> 66.40; Unique 0-20m mIoU 19.02 -> 18.93 | Mixed |
| apply DynaVG on 3EED Vehicle dataset | 3EED · 3EED Vehicle | Acc@0.5 IoU: 45.72 -> 46.97 | Improved |
Limitations
- Only the abstract was available; experimental details are unknown
Closest papers by meaning
- 3EED: Ground Everything Everywhere in 3D (NeurIPS 2025 · similarity 0.89)
- From Objects to Anywhere: A Holistic Benchmark for Multi-level Visual Grounding in 3D Scenes (NeurIPS 2025 · similarity 0.89)
- Visual Programming for Zero-shot Open-Vocabulary 3D Visual Grounding (CVPR 2024 · similarity 0.89)
- SeeGround: See and Ground for Zero-Shot Open-Vocabulary 3D Visual Grounding (CVPR 2025 · similarity 0.88)
- MolmoMotion: Forecasting Point Trajectories in 3D with Language Instruction (NeurIPS 2026 · similarity 0.88)
- Dynam3D: Dynamic Layered 3D Tokens Empower VLM for Vision-and-Language Navigation (NeurIPS 2025 · similarity 0.87)
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.
