Research map / Point clouds / Point cloud classification
JM3D & JM3D-LLM: Elevating 3D Representation With Joint Multi-Modal Cues
Jiayi Ji, Haowei Wang, Changli Wu, Yiwei Ma, Xiaoshuai Sun, Rongrong Ji
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 |
|---|---|---|---|
| use 4 randomly sampled views without embeddings instead of a single view (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (All) | top-1 accuracy: 60.0 -> 56.5 | Hurt |
| add angle and depth embeddings to multi-view images (ablation) | JM3D (PointMLP, 4 views) · ModelNet40 zero-shot (All) | top-1 accuracy: 56.5 -> 58.8 | Improved |
| sample views within an angular range instead of randomly (ablation) | JM3D (PointMLP, 4 views) · ModelNet40 zero-shot (All) | top-1 accuracy: 58.8 -> 60.7 | Improved |
| use 2 views instead of 4 views (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (All) | top-1 accuracy: 60.7 -> 61.2 | No clear effect |
| use subcategory text instead of parent category text (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (Hard) | top-1 accuracy: 51.4 -> 43.5 | Hurt |
| use subcategory text instead of parent category text (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (All) | top-1 accuracy: 61.2 -> 61.8 | No clear effect |
| add parent-category classification on top of subcategory text (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (All) | top-1 accuracy: 61.8 -> 63.1 | Improved |
| add JMA joint multi-modal alignment to CIS and HTT (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (All) | top-1 accuracy: 63.1 -> 65.8 | Improved |
| add JMA joint multi-modal alignment to CIS and HTT (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (Hard) | top-1 accuracy: 50.4 -> 55.0 | Improved |
| add JMA joint multi-modal alignment to CIS and HTT (ablation) | JM3D (PointMLP) · ScanObjectNN zero-shot (All) | top-1 accuracy: 46.9 -> 47.5 | Improved |
| add CIS and HTT without JMA to the baseline (ablation) | JM3D (PointMLP) · ModelNet40 zero-shot (All) | top-1 accuracy: 60.0 -> 63.1 | Improved |
| replace ULIP with JM3D | PointMLP · ModelNet40 zero-shot (All) | top-1 accuracy: 61.5 -> 65.8 | Improved |
Closest papers by meaning
- Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering (NeurIPS 2026 · similarity 0.89)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.89)
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · similarity 0.89)
- More Text, Less Point: Towards 3D Data-Efficient Point-Language Understanding (AAAI 2025 · similarity 0.88)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.88)
- TAMM: TriAdapter Multi-Modal Learning for 3D Shape Understanding (CVPR 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.
