Research map / Point clouds / Point cloud classification
Point Cloud Pre-Training With Natural 3D Structures
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Pre-train VoteNet on PC-FractalDB rather than train from scratch | VoteNet with PointNet++ · ScanNetV2 | mAP@0.25: 57.9 -> 61.9 | Helped |
| Pre-train VoteNet on PC-FractalDB rather than train from scratch | VoteNet with PointNet++ · SUN RGB-D | mAP@0.25: 57.4 -> 59.4 | Helped |
| Pre-train the full detection model instead of only the backbone through classification (ablation) | VoteNet · ScanNetV2 | mAP@0.25: 48.8 -> 61.1 | Helped |
| Use diverse fractal models, categories, and instances rather than one of each (ablation) | PC-FractalDB pre-training · ScanNetV2 | mAP@0.25: 57.2 -> 61.6 | Helped |
| Pre-train with fractal scenes instead of ModelNet CAD scenes | VoteNet · SUN RGB-D | mAP@0.25: 55.0 -> 58.0 | Helped |
| Add a variance threshold when defining fractal categories (ablation) | PC-FractalDB · ScanNetV2 | mAP@0.25: 58.9 -> 61.9 | Helped |
| Add FractalNoiseMix instance augmentation (ablation) | PC-FractalDB · SUN RGB-D | mAP@0.25: 57.5 -> 59.0 | Helped |
| Increase instances per category from 10 to 1,000 (ablation) | PC-FractalDB · SUN RGB-D | mAP@0.25: 58.2 -> 59.2 | Helped |
| Increase categories from 100 to 1,000 (ablation) | PC-FractalDB · ScanNetV2 | mAP@0.25: 61.0 -> 61.9 | Helped |
| Increase scenes from 1k to 10k (ablation) | PC-FractalDB · SUN RGB-D | mAP@0.25: 55.3 -> 59.0 | Helped |
| Use 15 rather than 5 objects per scene on average (ablation) | PC-FractalDB · ScanNetV2 | mAP@0.25: 59.4 -> 61.9 | Helped |
| Use formula-generated 3D IFS labels instead of self-supervised labels (ablation) | PC-FractalDB pre-training · SUN RGB-D | mAP@0.25: 54.3 -> 57.1 | Helped |
Closest papers by meaning
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.86)
- Digging into Intrinsic Contextual Information for High-fidelity 3D Point Cloud Completion (AAAI 2025 · similarity 0.85)
- X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition (CVPR 2024 · similarity 0.85)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.85)
- ComPC: Completing a 3D Point Cloud with 2D Diffusion Priors (ICLR 2025 · similarity 0.85)
- An Explicit Frame Construction for Normalizing 3D Point Clouds (ICML 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.