Research map / Point clouds / Point cloud classification
Any2Point: Empowering Any-modality Transformers for Efficient 3D Understanding
Yiwen Tang, Ray Zhang, Jiaming Liu, Zoey Guo, Bin Zhao, Zhigang Wang, Dong Wang, Peng Gao, Hongsheng Li, Xuelong Li
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 3D-to-any virtual projection positional encoding (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 88.7 -> 89.3 | Improved |
| add any-to-3D guided adapter (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 88.7 -> 90.9 | Improved |
| add both virtual projection and guided adapter (ablation) | Any2Point (DINO V2) · ScanObjectNN PB-T50-RS | accuracy (%): 86.1 -> 87.7 | Improved |
| use sinusoidal or learnable PE instead of no PE (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 90.9 -> 87.4 (sinusoidal), 90.5 (learnable) | Hurt |
| use virtual projection PE instead of learnable PE (ablation) | Any2Point (DINO V2) · ScanObjectNN PB-T50-RS | accuracy (%): 86.5 -> 87.7 | Improved |
| add 1D/2D-guided local aggregation to adapter (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 89.3 -> 90.2 | Improved |
| add adaptive any-to-3D ensemble (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 90.2 -> 91.9 | Improved |
| use guided adapter instead of LoRA (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 86.3 -> 91.9 | Improved |
| use guided adapter PEFT instead of full fine-tuning (ablation) | Any2Point (CLIP-Text) · ScanObjectNN PB-T50-RS | accuracy (%): 79.9 -> 91.9 | Improved |
| use language (1D) source model instead of 2D vision or audio | Any2Point · ScanObjectNN PB-T50-RS | accuracy (%): 87.7 -> 91.9 | Improved |
Closest papers by meaning
- Bridging 1D, 2D, and 3D with Any-to-Any Multimodal Modeling (NeurIPS 2026 · similarity 0.88)
- Inst3D-LMM: Instance-Aware 3D Scene Understanding with Multi-modal Instruction Tuning (CVPR 2025 · similarity 0.87)
- Unifying 2D and 3D Vision-Language Understanding (ICML 2025 · similarity 0.86)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.86)
- OneCanvas: 3D Scene Understanding via Panoramic Reprojection (NeurIPS 2026 · similarity 0.86)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 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.
