Research map / Detection and segmentation / CNN features
End-to-End 3D Dense Captioning With Vote2Cap-DETR
Sijin Chen, Hongyuan Zhu, Xin Chen, Yinjie Lei, Gang Yu, 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 |
|---|---|---|---|
| Use one-stage Vote2Cap-DETR instead of the strongest listed two-stage baseline | 3D dense captioning · ScanRefer val, IoU=0.50, SCST, w/ additional 2D | CIDEr: 62.64 -> 70.63 | Improved |
| Use one-stage Vote2Cap-DETR instead of the strongest listed baseline | 3D dense captioning · Nr3D, IoU=0.50, SCST | CIDEr: 38.42 -> 45.53 | Improved |
| Add vote query positions instead of using seed positions (ablation) | 3DETR object detection · ScanNet detection, IoU=0.50 | mAP: 48.18 -> 49.15 | Improved |
| Add gathered content features to the vote query (ablation) | Vote-position 3DETR query · ScanNet detection, IoU=0.50 | mAP: 49.15 -> 52.13 | Improved |
| Add gathered content features to the vote query (ablation) | Vote-position 3DETR query · ScanNet detection, first decoder layer, IoU=0.50 | mAP: 30.23 -> 46.53 | Improved |
| Add local contextual features to the caption decoder (ablation) | Object-centric caption decoder · ScanRefer, IoU=0.50, MLE | CIDEr: 60.15 -> 61.39 | Improved |
| Use local rather than global scene context tokens (ablation) | Caption decoder with contextual features · ScanRefer, IoU=0.50, MLE | CIDEr: 61.20 -> 61.39 | Improved |
| Use set-to-set training instead of sentence training (ablation) | Vote2Cap-DETR · ScanRefer, IoU=0.50, SCST | CIDEr: 71.39 -> 73.77 | Improved |
| Use set-to-set training instead of sentence training (ablation) | Vote2Cap-DETR · ScanRefer, IoU=0.50, MLE | CIDEr: 61.21 -> 61.81 | Improved |
| Remove NMS (ablation) | Vote2Cap-DETR · IoU=0.50 | CIDEr: 70.63 -> 71.57 | Improved |
| Remove NMS (ablation) | Vote2Cap-DETR · IoU=0.50 | mAP50: 52.79 -> 52.82 | No clear effect |
| Replace the scene encoder, vote query, and transformer decoder with VoteNet (ablation) | Vote2Cap-DETR · ScanRefer, IoU=0.50, MLE | CIDEr: 59.32 -> 52.96 | Hurt |
Closest papers by meaning
- Vote2Cap-DETR++: Decoupling Localization and Describing for End-to-End 3D Dense Captioning (TPAMI 2024 · similarity 0.94)
- Scan2Cap: Context-Aware Dense Captioning in RGB-D Scans (CVPR 2021 · similarity 0.89)
- DetCLIPv3: Towards Versatile Generative Open-vocabulary Object Detection (CVPR 2024 · similarity 0.87)
- Dense Video Object Captioning from Disjoint Supervision (ICLR 2025 · similarity 0.86)
- A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions (CVPR 2024 · similarity 0.86)
- Streaming Dense Video Captioning (CVPR 2024 · similarity 0.85)
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.
