Research map / Point clouds / Point cloud classification
PointCLIP V2: Prompting CLIP and GPT for Powerful 3D Open-world Learning
Xiangyang Zhu, Renrui Zhang, Bowei He, Ziyu Guo, Ziyao Zeng, Zipeng Qin, Shanghang Zhang, Peng Gao
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 GPT-3 generated 3D-specific text prompts (ablation) | PointCLIP V2 · ModelNet40 zero-shot | accuracy (%): 39.11 -> 64.22 | Improved |
| add densify step (local min pooling) (ablation) | PointCLIP V2 projection · ModelNet40 zero-shot | accuracy (%): +15.14% | Improved |
| add Gaussian smoothing step (ablation) | PointCLIP V2 projection · ModelNet40 zero-shot | accuracy (%): 59.64 -> 64.22 | Improved |
| add quantize (voxel grid) step instead of orthogonal projection (ablation) | PointCLIP V2 projection · ModelNet40 zero-shot | accuracy (%): 57.35 -> 64.22 | Improved |
| use min pooling instead of max or average pooling for densify (ablation) | PointCLIP V2 projection · ModelNet40 zero-shot | accuracy (%): 57.35 (max), 60.71 (avg) -> 64.22 | Improved |
| add question-answering command to caption command (ablation) | GPT prompting · ModelNet40 zero-shot | accuracy (%): 61.67 -> 60.86 | No clear effect |
| use all four GPT command types instead of caption only (ablation) | GPT prompting · ModelNet40 zero-shot | accuracy (%): 61.67 -> 64.22 | Improved |
| make smoothing learnable with 3D conv layers in few-shot setting (ablation) | PointCLIP V2 few-shot · ModelNet40 16-shot | accuracy (%): +0.7% | Improved |
| add 3D-specific text in few-shot setting (ablation) | PointCLIP V2 few-shot · 16-shot | accuracy (%): +3.33% | Improved |
| increase sampled points from 1024 to 8192 (ablation) | PointCLIP V2 · ModelNet40 zero-shot | accuracy (%): 64.22 -> 68.56 | Improved |
| use ViT-B/16 backbone instead of other CLIP visual encoders (ablation) | PointCLIP V2 · ModelNet40 zero-shot | accuracy (%): 46.45 (RN50) -> 64.22 | Improved |
Later papers that compared with this method
8 later accepted papers report PointCLIPV2 as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.
- Back to Point: Exploring Point-Language Models for Zero-Shot 3D Anomaly Detection (CVPR 2026 · better on 2 of 2 shared results)
- PatchAlign3D: Local Feature Alignment for Dense 3D Shape Understanding (CVPR 2026 · better on 2 of 2 shared results)
- Geometry-Centered 3D Latent World Models for Growing Surfaces (NeurIPS 2026 · better on 1 of 1 shared result)
- CLIPoint3D: Language-Grounded Few-Shot Unsupervised 3D Point Cloud Domain Adaptation (CVPR 2026 · better on 1 of 1 shared result)
- Point Cloud Quantization Through Multimodal Prompting for 3D Understanding (AAAI 2026 · better on 1 of 1 shared result)
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · better on 2 of 2 shared results)
- OpenDlign: Open-World Point Cloud Understanding with Depth-Aligned Images (NeurIPS 2024 · better on 2 of 2 shared results)
- OpenShape: Scaling Up 3D Shape Representation Towards Open-World Understanding (NeurIPS 2023 · better on 2 of 2 shared results)
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.
- CLIP2Point — CLIP2Point: Transfer CLIP to Point Cloud Classification with Image-Depth Pre-Training (ICCV 2023) · better on 2 of 2 shared results
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
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · similarity 0.90)
- Occlusion-aware Text-Image-Point Cloud Pretraining for Open-World 3D Object Recognition (CVPR 2025 · similarity 0.89)
- Towards 3D Objectness Learning in an Open World (NeurIPS 2025 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.88)
- OpenDlign: Open-World Point Cloud Understanding with Depth-Aligned Images (NeurIPS 2024 · similarity 0.88)
- RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding (CVPR 2024 · similarity 0.87)
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.
