Research map / Point clouds / Point cloud classification
Invariant Training 2D-3D Joint Hard Samples for Few-Shot Point Cloud Recognition
Xuanyu Yi, Jiajun Deng, Qianru Sun, Xian-Sheng Hua, Joo-Hwee Lim, Hanwang Zhang
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 |
|---|---|---|---|
| apply InvJoint training instead of PointCLIP | InvJoint · ModelNet40 1-shot | accuracy (%): 52.96 -> 68.85 | Improved |
| naive late-fusion ensemble of 2D PointCLIP and 3D DGCNN | PointCLIP · ModelNet40 1-shot | accuracy (%): 52.96 -> 53.70 | No clear effect |
| add joint hard sample selection (Step 1) (ablation) | InvJoint · ModelNet40 16-shot | accuracy (%): 83.91 -> 88.97 | Improved |
| add cross-modal invariance learning (Step 2) (ablation) | InvJoint · ModelNet40 16-shot | accuracy (%): 85.82 -> 88.97 | Improved |
| add both Step 1 and Step 2 over plain CE training with product fusion (ablation) | InvJoint · ModelNet40 16-shot | accuracy (%): 81.09 -> 88.97 | Improved |
| use product fusion instead of sum fusion (ablation) | InvJoint · ModelNet40 16-shot | accuracy (%): 87.61 -> 88.97 | Improved |
| use learnable prompt tokens instead of handcrafted templates (ablation) | InvJoint · Toys4K 16-shot | accuracy (%): 93.52 -> 78.57 | Hurt |
| joint prediction reduces sensitivity to prompt design (ablation) | InvJoint · Toys4K 16-shot | std across prompts (%): 10.87 -> 5.51 | Improved |
| use ViT-B/16 instead of RN50 visual encoder (ablation) | InvJoint · ModelNet40 16-shot | accuracy (%): 85.93 -> 88.97 | Improved |
| use InvJoint instead of plain late fusion (ablation) | InvJoint · ModelNet40 | accuracy (%): +4.7% on average | Improved |
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.
- Crosspoint — CrossPoint: Self-Supervised Cross-Modal Contrastive Learning for 3D Point Cloud Understanding (CVPR 2022) · better on 1 of 1 shared result
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
- A Closer Look at Few-Shot 3D Point Cloud Classification (IJCV 2022 · similarity 0.89)
- Voint Cloud: Multi-View Point Cloud Representation for 3D Understanding (ICLR 2023 · similarity 0.87)
- Zero-Shot Learning on 3D Point Cloud Objects and Beyond (IJCV 2022 · similarity 0.86)
- 3D-ANC: Adaptive Neural Collapse for Robust 3D Point Cloud Recognition (AAAI 2026 · similarity 0.86)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.86)
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 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.
