Research map / 3D vision / 3D generation
RGB-Pointmap Pretraining for Unified 3D Scene Understanding
Ye Mao, Weixun Luo, Ranran Huang, Junpeng Jing, Krystian Mikolajczyk
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 pointmap input to image-only encoder (ablation) | UniScene3D · ScanRefer viewpoint grounding | R@1: 28.8 -> 38.6 | Improved |
| add image input to pointmap-only encoder (ablation) | UniScene3D · ScanRefer viewpoint grounding | R@1: 37.6 -> 38.6 | Improved |
| add image input to pointmap-only encoder (ablation) | UniScene3D · ScanNet scene type classification (0-shot) | accuracy: 67.2 -> 70.7 | Improved |
| add image input to pointmap-only encoder (ablation) | UniScene3D · ScanNet scene type classification (10-shot) | accuracy: 83.6 -> 83.7 | No clear effect |
| add grounded view alignment loss (ablation) | UniScene3D · ScanRefer viewpoint grounding | R@1: 18.5 -> 38.6 | Improved |
| add grounded view alignment loss (ablation) | UniScene3D · Sr3D viewpoint grounding | R@1: 11.5 -> 23.6 | Improved |
| add cross-view geometric alignment loss (ablation) | UniScene3D · ScanRefer viewpoint grounding | R@1: 38.2 -> 38.6 | Improved |
| add cross-view geometric alignment loss (ablation) | UniScene3D · Nr3D viewpoint grounding | R@1: 23.2 -> 25.2 | Improved |
| add cross-view geometric alignment loss (ablation) | UniScene3D · ScanNet scene type classification (10-shot) | accuracy: 81.3 -> 83.7 | Improved |
| add grounded view alignment loss (ablation) | UniScene3D · ScanRefer scene retrieval (n=10) | R@5: 63.0 -> 62.9 | Hurt |
| use summation early fusion instead of concat + projection (ablation) | UniScene3D · ScanRefer viewpoint grounding | R@1: 36.5 -> 38.6 | Improved |
| initialize pointmap patch embedding from image patch embedding (ablation) | UniScene3D · ScanRefer viewpoint grounding | R@1: 38.0 -> 38.6 | Improved |
Closest papers by meaning
- Uni3D: Exploring Unified 3D Representation at Scale (ICLR 2024 · similarity 0.89)
- A Unified Framework for 3D Scene Understanding (NeurIPS 2024 · similarity 0.89)
- UniGS: Unified Language-Image-3D Pretraining with Gaussian Splatting (ICLR 2025 · similarity 0.88)
- Building a Strong Pre-Training Baseline for Universal 3D Large-Scale Perception (CVPR 2024 · similarity 0.88)
- CLIP-Driven Open-Vocabulary 3D Scene Graph Generation via Cross-Modality Contrastive Learning (CVPR 2024 · similarity 0.88)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.88)
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.
