Research map / Point clouds / Segmentation and detection
SegDINO3D: 3D Instance Segmentation Empowered by Both Image-Level and Object-Level 2D Features
Jinyuan Qu, Hongyang Li, Xingyu Chen, Shilong Liu, Yukai Shi, Tianhe Ren, Ruitao Jing, Lei Zhang
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 globally enhanced 2D image-level features in the encoder (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 26.8 -> 35.2 | Improved |
| add 2D object-level queries in the decoder (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 26.8 -> 33.0 | Improved |
| combine object-level features with image-level features (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 35.2 -> 38.3 | Improved |
| add box-modulated cross-attention (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 38.3 -> 39.8 | Improved |
| use predicted boxes to filter distant masks (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 39.8 -> 40.2 | Improved |
| use 3D encoder global context fusion instead of MLP fusion (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 35.9 -> 39.8 | Improved |
| use nearest views instead of farthest views (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 36.2 -> 39.8 | Improved |
| add distance-aware attention mask in DACA-2D (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 36.3 -> 39.8 | Improved |
| use 2D object queries instead of mapping 2D features to 3D points (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 35.7 -> 39.8 | Improved |
| add 70% dropout of 2D object queries in training (ablation) | SegDINO3D · ScanNet200 val (DINO-X ablation) | mAP: 39.2 -> 39.8 | Improved |
| add center positional embedding without box modulation (ablation) | SegDINO3D box queries · ScanNet200 val (DINO-X ablation) | mAP: 38.3 -> 38.1 | No clear effect |
| add box size regression without modulation (ablation) | SegDINO3D box queries · ScanNet200 val (DINO-X ablation) | mAP: 38.1 -> 37.8 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Relation3D — Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025) · better on 2 of 2 shared results
- OneFormer3D — OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024) · better on 2 of 2 shared results
- Spherical Mask — Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation (CVPR 2024) · better on 1 of 1 shared result
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
- SA3DIP: Segment Any 3D Instance with Potential 3D Priors (NeurIPS 2024 · similarity 0.88)
- Open-YOLO 3D: Towards Fast and Accurate Open-Vocabulary 3D Instance Segmentation (ICLR 2025 · similarity 0.87)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.87)
- Seg2Box: 3D Object Detection by Point-Wise Semantics Supervision (AAAI 2025 · similarity 0.87)
- ODIN: A Single Model for 2D and 3D Segmentation (CVPR 2024 · similarity 0.87)
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · 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.
