Research map / Vision-language and multimodal / Human and embodied modeling
D3Net: A Unified Speaker-Listener Architecture for 3D Dense Captioning and Visual Grounding
Zhenyu Chen, Qirui Wu, Matthias Nießner, 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 |
|---|---|---|---|
| train speaker with CIDEr reward (REINFORCE) instead of MLE (ablation) | D3Net · ScanRefer dense captioning (val) | CIDEr@0.5IoU: 46.07 -> 57.88 | Improved |
| add trainable localization listener reward (ablation) | D3Net (CIDEr) · ScanRefer dense captioning (val) | CIDEr@0.5IoU: 57.88 -> 61.30 | Improved |
| use frozen pretrained listener instead of jointly fine-tuned listener (ablation) | D3Net (CIDEr+loc.) · ScanRefer dense captioning (val) | CIDEr@0.5IoU: 61.30 -> 58.93 | Hurt |
| add language object classification loss to listener reward (ablation) | D3Net (CIDEr+loc.) · ScanRefer dense captioning (val) | CIDEr@0.5IoU: 61.30 -> 61.50 | Improved |
| add CIDEr reward (ablation) | D3Net · ScanRefer dense captioning (val) | mAP@0.5: 50.93 -> 51.01 | No clear effect |
| add listener rewards (loc. + lobjcls.) (ablation) | D3Net (CIDEr) · ScanRefer dense captioning (val) | mAP@0.5: 51.01 -> 52.58 | Improved |
| add 1x extra unannotated ScanNet scans (ablation) | D3Net (CIDEr+loc.+lobjcls.) · ScanRefer dense captioning (val) | CIDEr@0.5IoU: 61.50 -> 62.64 | Improved |
| add 1x extra unannotated ScanNet scans (ablation) | D3Net (CIDEr+loc.+lobjcls.) · ScanRefer dense captioning (val) | mAP@0.5: 52.58 -> 53.95 | Improved |
| switch detector from VoteNet to PointGroup (ablation) | D3Net (CIDEr+lis.+extra) · ScanRefer dense captioning (val) | CIDEr@0.5IoU: 48.38 -> 62.64 | Improved |
| fine-tune jointly with speaker (speaker-listener) after pretrained listener (ablation) | D3Net · ScanRefer visual grounding (val) | Overall Acc@0.5IoU: 35.58 -> 37.87 | Improved |
| fine-tune jointly with speaker (ablation) | D3Net · ScanRefer visual grounding (val, Unique subset) | Acc@0.5IoU: 70.35 -> 72.04 | Mixed |
| fine-tune jointly with speaker (ablation) | D3Net · ScanRefer visual grounding (test) | Overall Acc@0.5IoU: 35.90 -> 39.19 | Improved |
Later papers that compared with this method
1 later accepted paper reports D3Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- UniT3D: A Unified Transformer for 3D Dense Captioning and Visual Grounding (ICCV 2023 · better on 5 of 5 shared results)
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
- UniT3D: A Unified Transformer for 3D Dense Captioning and Visual Grounding (ICCV 2023 · similarity 0.90)
- From Thousands to Billions: 3D Visual Language Grounding via Render-Supervised Distillation from 2D VLMs (ICML 2025 · similarity 0.88)
- Unifying 2D and 3D Vision-Language Understanding (ICML 2025 · similarity 0.87)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.87)
- DenseGrounding: Improving Dense Language-Vision Semantics for Ego-centric 3D Visual Grounding (ICLR 2025 · similarity 0.87)
- Towards CLIP-driven Language-free 3D Visual Grounding via 2D-3D Relational Enhancement and Consistency (CVPR 2024 · 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.
