Research map / Vision-language and multimodal / Human and embodied modeling
Multi3DRefer: Grounding Text Description to Multiple 3D Objects
Yiming Zhang, ZeMing Gong, Angel X. Chang
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 contrastive loss between sentence and object features (ablation) | M3DRef-CLIP · Nr3D val (GT boxes) | accuracy All: 42.9 -> 49.4 | Improved |
| add contrastive loss (ablation) | M3DRef-CLIP · Multi3DRefer val (GT boxes) | F1 overall: 61.1 -> 62.3 | Improved |
| add contrastive loss (ablation) | M3DRef-CLIP · ScanRefer val (GT boxes) | accuracy All: 55.4 -> 56.9 | Improved |
| replace GRU+GloVe text encoder with CLIP text encoder (ablation) | M3DRef-CLIP · Multi3DRefer val (GT boxes) | F1 overall: 59.1 -> 61.3 | Improved |
| add CLIP image features of rendered proposals to 3D features (ablation) | M3DRef-CLIP · Multi3DRefer val (GT boxes) | F1 overall: 61.3 -> 62.3 | Improved |
| use CLIP image features alone instead of 3D features (ablation) | M3DRef-CLIP · Multi3DRefer val (GT boxes) | F1 overall: 61.3 -> 54.4 | Hurt |
| train on Multi3DRefer (ST) plus ScanRefer (ablation) | M3DRef-CLIP · ScanRefer val (GT boxes) | accuracy All: 56.9 -> 58.7 | Improved |
| train on Multi3DRefer (ST) data only versus ScanRefer (ablation) | M3DRef-CLIP · ScanRefer val (GT boxes) | accuracy All: 56.9 -> 55.0 | Hurt |
| M3DRef-CLIP versus prior methods on ScanRefer | M3DRef-CLIP · ScanRefer val Acc@0.5 | accuracy All: 40.6 -> 44.7 | Improved |
| M3DRef-CLIP on Nr3D with GT boxes | M3DRef-CLIP · Nr3D val | accuracy All: 55.4 (MVT) vs 49.4 | Hurt |
| M3DRef-CLIP versus 3DJCG on zero-target handling with predicted boxes | M3DRef-CLIP · Multi3DRefer ZT w/ D (pred boxes) | F1@0.5: 66.9 (3DJCG) vs 39.4 | Hurt |
Closest papers by meaning
- From Objects to Anywhere: A Holistic Benchmark for Multi-level Visual Grounding in 3D Scenes (NeurIPS 2025 · similarity 0.91)
- AugRefer: Advancing 3D Visual Grounding via Cross-Modal Augmentation and Spatial Relation-based Referring (AAAI 2025 · similarity 0.90)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.90)
- G^3-LQ: Marrying Hyperbolic Alignment with Explicit Semantic-Geometric Modeling for 3D Visual Grounding (CVPR 2024 · similarity 0.90)
- MiKASA: Multi-Key-Anchor & Scene-Aware Transformer for 3D Visual Grounding (CVPR 2024 · similarity 0.89)
- CoT3DRef: Chain-of-Thoughts Data-Efficient 3D Visual Grounding (ICLR 2024 · similarity 0.89)
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.
