Research map / Point clouds / Point cloud classification
3DJCG: A Unified Framework for Joint Dense Captioning and Visual Grounding on 3D Point Clouds
Daigang Cai, Lichen Zhao, Jing Zhang, Lu Sheng, Dong Xu
Check 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 |
|---|---|---|---|
| Jointly train captioning and grounding instead of removing the grounding head (ablation) | 3DJCG · Scan2Cap | B-4@0.5: 26.24 -> 31.03 | Improved |
| Add attribute and relation-aware feature enhancement (ablation) | 3DJCG · Scan2Cap | B-4@0.5: 29.08 -> 31.03 | Improved |
| Add ScanRefer grounding data to the training data (ablation) | 3DJCG · Scan2Cap | B-4@0.5: 30.40 -> 31.03 | Improved |
| Jointly train captioning and grounding instead of removing the captioning head (ablation) | 3DJCG · ScanRefer | Overall Acc@0.5: 36.72 -> 37.33 | Improved |
| Add attribute and relation-aware feature enhancement (ablation) | 3DJCG · ScanRefer | Overall Acc@0.5: 35.12 -> 37.33 | Improved |
| Add Scan2Cap captioning data to the training data (ablation) | 3DJCG · ScanRefer | Overall Acc@0.5: 36.93 -> 37.33 | Improved |
| Fine-tune the pretrained captioning model on Nr3D (ablation) | 3DJCG-C · Nr3D | B-4@0.5: 20.45 -> 22.82 | Improved |
| Use 3DJCG-C from scratch rather than Scan2Cap | 3D captioning · Nr3D | B-4@0.5: 17.24 -> 20.45 | Improved |
| Use 3DJCG instead of 3DVG-Transformer | 3D visual grounding · ScanRefer validation, 2D + 3D | Overall Acc@0.5: 34.67 -> 37.33 | Improved |
| Use 3DJCG instead of Scan2Cap | 3D dense captioning · Scan2Cap validation, 2D + 3D | C@0.5: 39.08 -> 49.48 | Improved |
Later papers that compared with this method
1 later accepted paper reports 3DJCG as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- WildRefer: 3D Object Localization in Large-scale Dynamic Scenes with Multi-modal Visual Data and Natural Language (ECCV 2024 · better on 1 of 1 shared result)
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.
- Scan2Cap — Scan2Cap: Context-Aware Dense Captioning in RGB-D Scans (CVPR 2021) · better on 8 of 8 shared results
- 3DVG-Transformer — 3DVG-Transformer: Relation Modeling for Visual Grounding on Point Clouds (ICCV 2021) · better on 6 of 6 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
- From Objects to Anywhere: A Holistic Benchmark for Multi-level Visual Grounding in 3D Scenes (NeurIPS 2025 · similarity 0.89)
- Towards CLIP-driven Language-free 3D Visual Grounding via 2D-3D Relational Enhancement and Consistency (CVPR 2024 · similarity 0.88)
- DenseGrounding: Improving Dense Language-Vision Semantics for Ego-centric 3D Visual Grounding (ICLR 2025 · similarity 0.88)
- Cycle-Consistency Learning for Captioning and Grounding (AAAI 2024 · similarity 0.88)
- Grounded 3D-Aware Spatial Vision-Language Modeling (CVPR 2026 · similarity 0.88)
- Multi-Object 3D Grounding with Dynamic Modules and Language-Informed Spatial Attention (NeurIPS 2024 · 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.
