Research map / Point clouds / Point cloud classification
ORACLE-3D: Open-world Region-aligned Cross-modal Learning for Label-efficient 3D Scene Understanding
Yuru Wang, Pei Liu, Songtao Wang, Zehan Zhang, Xinyan Lu, Changwei Cai, Hao Li, Haipeng LIU, Qingtian Ning, Jun Ma
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 two-stage training instead of one-stage with logit loss only (ablation) | ORACLE-3D · nuScenes B12/N3 | mIoU base: 73.2 -> 76.6 | Improved |
| use two-stage training instead of one-stage with logit loss only (ablation) | ORACLE-3D · nuScenes B12/N3 | mIoU novel: 57.3 -> 60.1 | Improved |
| use two-stage training with all losses instead of one-stage (ablation) | ORACLE-3D · nuScenes B12/N3 | mIoU novel: 58.2 -> 66.5 | Improved |
| add novel-class logit distillation loss (ablation) | ORACLE-3D two-stage · nuScenes B12/N3 | mIoU novel: 60.1 -> 62.6 | Improved |
| add feature distillation (ablation) | ORACLE-3D two-stage · nuScenes B12/N3 | mIoU novel: 62.6 -> 65.2 | Improved |
| add vision-point matching loss (ablation) | ORACLE-3D two-stage · nuScenes B12/N3 | mIoU novel: 65.2 -> 66.5 | Improved |
| add vision-point matching loss (ablation) | ORACLE-3D two-stage · nuScenes B12/N3 | mIoU base: 76.8 -> 76.9 | Improved |
| add distillation losses and VPM to two-stage logit-only model (ablation) | ORACLE-3D · nuScenes B12/N3 | mIoU base: 76.6 -> 76.9 | No clear effect |
| replace 2D-3D projection with full ORACLE-3D framework | annotation-free nuScenes · nuScenes annotation-free | mIoU: 28.6 -> 56.9 | Improved |
| replace RegionPLC with ORACLE-3D on same SparseUNet32 | nuScenes · nuScenes B10/N5 | mIoU novel B10/N5: 36.3 -> 57.3 | Improved |
| swap SparseUNet32 backbone for PTv3 (ablation) | ORACLE-3D · nuScenes B12/N3 | mIoU novel: 62.2 -> 66.5 | Improved |
| replace RegionPLC with ORACLE-3D on ScanNet | SparseUNet32 · ScanNet B10/N9 | mIoU novel B10/N9: 55.6 -> 67.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CLIP2Scene — CLIP2Scene: Towards Label-Efficient 3D Scene Understanding by CLIP (CVPR 2023) · 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
- CrossOver: 3D Scene Cross-Modal Alignment (CVPR 2025 · similarity 0.91)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.91)
- Inst3D-LMM: Instance-Aware 3D Scene Understanding with Multi-modal Instruction Tuning (CVPR 2025 · similarity 0.91)
- OpenDlign: Open-World Point Cloud Understanding with Depth-Aligned Images (NeurIPS 2024 · similarity 0.90)
- RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding (CVPR 2024 · similarity 0.90)
- Hierarchical Cross-Modal Alignment for Open-Vocabulary 3D Object Detection (AAAI 2025 · similarity 0.90)
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.
