Research map / Detection and segmentation / 3D detection and driving
Vote2Cap-DETR++: Decoupling Localization and Describing for End-to-End 3D Dense Captioning
Sijin Chen, Hongyuan Zhu, Mingsheng Li, Xin Chen, Peng Guo, Yinjie Lei, Gang Yu, Taihao Li, Tao Chen
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 vote query position p_vq as query position (ablation) | 3DETR-m · ScanNet val | mAP@0.5: +0.97% mAP@0.5 | Improved |
| add aggregated local feature f_vq as initial query feature (ablation) | Vote2Cap-DETR · ScanNet val | mAP@0.5: +2.98% mAP@0.5 | Improved |
| add local context tokens (DCC) in caption head (ablation) | object-centric decoder · ScanRefer val (frozen backbone) | C@0.5: 60.15 -> 61.39 | Improved |
| use set-to-set training instead of sentence training under SCST (ablation) | Vote2Cap-DETR · ScanRefer val | C@0.5: 71.39 -> 73.77 | Improved |
| use set-to-set training under MLE (ablation) | Vote2Cap-DETR · ScanRefer val | C@0.5: 61.21 -> 61.81 | No clear effect |
| add detection pretraining before end-to-end training (ablation) | Vote2Cap-DETR · ScanRefer val | C@0.5: 52.15 -> 62.03 | Improved |
| remove NMS at inference (ablation) | Vote2Cap-DETR · ScanRefer val | C@0.5: 70.63 -> 71.57 | No clear effect |
| remove NMS at inference (ablation) | SpaCap3D · ScanRefer val | mAP50: 37.77 -> 23.30 | Hurt |
| refine vote queries spatially in first three decoder layers (ablation) | Vote2Cap-DETR · ScanNet val | mAP@0.5: 52.13 -> 55.48 | Improved |
| refine vote queries in all decoder layers instead of first three (ablation) | Vote2Cap-DETR++ · ScanNet val | mAP@0.5: 55.48 -> 54.55 | Hurt |
| decouple loc and caption queries without correspondence (ablation) | Vote2Cap-DETR · ScanRefer val | C@0.5: 66.01 -> 65.38 | Hurt |
| link caption queries with loc queries through token-wise projection (ablation) | decoupled queries · ScanRefer val | C@0.5: 65.38 -> 67.58 | Improved |
Later papers that compared with this method
1 later accepted paper reports Vote2Cap-DETR as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- MTA: Multimodal Task Alignment for BEV Perception and Captioning (CVPR 2026 · better on 2 of 2 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
- End-to-End 3D Dense Captioning With Vote2Cap-DETR (CVPR 2023 · similarity 0.94)
- TOD3Cap: Towards 3D Dense Captioning in Outdoor Scenes (ECCV 2024 · similarity 0.91)
- Scan2Cap: Context-Aware Dense Captioning in RGB-D Scans (CVPR 2021 · similarity 0.90)
- Dense Video Object Captioning from Disjoint Supervision (ICLR 2025 · similarity 0.88)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.87)
- Chat-Scene: Bridging 3D Scene and Large Language Models with Object Identifiers (NeurIPS 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.
