Research map / Point clouds / Point cloud classification
3D sans 3D Scans: Scalable Pre-training from Video-Generated Point Clouds
Ryousuke Yamada, Kohsuke Ide, Yoshihiro Fukuhara, Hirokatsu Kataoka, Gilles Puy, Andrei Bursuc, Yuki M. Asano
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 Z-axis alignment to RoomTours (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 51.8 -> 57.1 | Improved |
| add scale alignment to ScanNet scale distribution (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 55.5 -> 57.1 | Improved |
| add noise-regularized loss (Laplacian + consistency) (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 51.1 -> 57.1 | Improved |
| add Laplacian smoothing loss (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 55.0 -> 57.1 | Improved |
| add noise consistency loss (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 55.3 -> 57.1 | Improved |
| scale RoomTours from 1k to 49k scenes | LAM3C · ScanNet semantic seg (LP) | mIoU: 57.1 -> 65.9 | Improved |
| use pi3 instead of VGGT for reconstruction (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 49.3 -> 57.1 | Improved |
| use pi3 instead of MapAnything (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 50.1 -> 57.1 | Improved |
| use LAM3C instead of Sonata on clean ScanNet pre-training | pre-training on ScanNet · ScanNet semantic seg (LP) | mIoU: 67.1 vs 66.8 | No clear effect |
| use LAM3C instead of Sonata on noisy RoomTours-1k | pre-training on RoomTours · ScanNet semantic seg (LP) | mIoU: 51.1 -> 57.1 | Improved |
| include val/test scenes in Sonata pre-training (ablation) | Sonata (ScanNet) · ScanNet semantic seg (Full-FT) | mIoU: 75.4 -> 77.4 | Improved |
| kNN size k=24 vs k=8 for Laplacian graph (ablation) | LAM3C · ScanNet semantic seg (LP) | mIoU: 56.0 -> 57.1 | Mixed |
Closest papers by meaning
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.88)
- Self-Supervised Learning on 3D Point Clouds by Learning Discrete Generative Models (CVPR 2021 · similarity 0.87)
- Spatio-Temporal Self-Supervised Representation Learning for 3D Point Clouds (ICCV 2021 · similarity 0.87)
- Self-Supervised Pretraining for Large-Scale Point Clouds (NeurIPS 2022 · similarity 0.87)
- UnScene3D: Unsupervised 3D Instance Segmentation for Indoor Scenes (CVPR 2024 · similarity 0.86)
- Generating 3D-Consistent Videos from Unposed Internet Photos (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.
