Research map / Point clouds / Point cloud classification
P2P: Tuning Pre-trained Image Models for Point Cloud Analysis with Point-to-Pixel Prompting
Read the paperCheck 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 |
|---|---|---|---|
| use Point-to-Pixel prompting instead of fine-tuning all ViT blocks (ablation) | ViT-B · ModelNet40 (ViT-B-1k) | accuracy: 90.6 -> 92.7 | Helped |
| use P2P instead of deep visual prompt tuning (ablation) | ViT-B · ModelNet40 (ViT-B-1k) | accuracy: 90.0 -> 92.7 | Helped |
| add pre-trained image model via P2P to geometry encoder (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 88.5 -> 92.7 | Helped |
| predict colors (colorful image) instead of binary image (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 89.8 -> 92.7 | Helped |
| generate explicit image instead of patch-embedding pixel features (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 90.8 -> 92.7 | Helped |
| use pre-trained weights instead of training ViT from scratch (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 86.3 -> 91.7 | Helped |
| tune normalization params instead of all params (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 91.7 -> 92.7 | Helped |
| tune normalization params instead of freezing or bias-only (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 92.2 -> 92.7 | Helped |
| use supervised IN-1k pre-training instead of MAE (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 91.6 -> 92.7 | Helped |
| use DINO instead of supervised IN-1k pre-training (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 92.7 -> 92.8 | No clear effect |
| increase image model scale (ViT-T to ViT-B) | P2P · ScanObjectNN PB-T50-RS | accuracy: 79.7 -> 83.4 | Helped |
| increase test views from 2 to 10 theta values (ablation) | P2P · ModelNet40 (ViT-B-1k) | accuracy: 90.2 -> 92.7 | Helped |
Closest papers by meaning
- GAPrompt: Geometry-Aware Point Cloud Prompt for 3D Vision Model (ICML 2025 · similarity 0.91)
- Point-PEFT: Parameter-Efficient Fine-Tuning for 3D Pre-trained Models (AAAI 2024 · similarity 0.90)
- Exploring Vision Semantic Prompt for Efficient Point Cloud Understanding (ICML 2025 · similarity 0.90)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.89)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.89)
- Point-PRC: A Prompt Learning Based Regulation Framework for Generalizable Point Cloud Analysis (NeurIPS 2024 · 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.