Research map / Point clouds / Point cloud classification
Efficient Encoder-Free Fourier-based 3D Large Multimodal Model
Guofeng Mei, Wei Lin, Luigi Riz, Yujiao Wu, Yiming Wang, Fabio Poiesi
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 superpoint pooling to point tokens (ablation) | Fase3D tokenizer · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 76.04 -> 79.70 | Improved |
| add FFT context enhancer to point tokens (ablation) | Fase3D tokenizer · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 76.04 -> 82.97 | Improved |
| add both superpoint pooling and FFT enhancer (ablation) | Fase3D tokenizer · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 76.04 -> 86.91 | Improved |
| decouple LoRA per branch instead of shared LoRA (ablation) | Fase3D LoRA · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 78.24 -> 82.53 | Improved |
| add Fourier residual to vision-branch LoRA (ablation) | dLoRA · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 82.53 -> 86.91 | Improved |
| also add Fourier residual to text-branch LoRA (ablation) | dLoRA+FFT · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 86.91 -> 83.64 | Hurt |
| replace Q-Former pooling with graph-based token merging (ablation) | token selector · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 77.41 -> 86.91 | Improved |
| replace FPS with graph-based token merging (ablation) | token selector · ScanQA (val, from scratch, Qwen2.5-3B) | CIDEr: 82.54 -> 86.91 | Improved |
| add two-stage generalist pretraining (ablation) | Fase3D · ScanQA (val) | CIDEr: 86.91 -> 91.74 | Improved |
| increase LoRA layers from 0 to 8 (ablation) | Fase3D LLM · ScanQA (val) | CIDEr: 74.03 -> 86.91 | Improved |
| increase LoRA layers from 10 to 12 (ablation) | Fase3D LLM · ScanQA (val) | CIDEr: 87.05 -> 86.40 | Hurt |
| increase space-filling curves from 1 to 4 (ablation) | Fase3D serialization · ScanQA (val) | CIDEr: 79.06 -> 86.91 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PerLA — PerLA: Perceptive 3D Language Assistant (CVPR 2025) · better on 4 of 5 shared results
- LL3DA — LL3DA: Visual Interactive Instruction Tuning for Omni-3D Understanding Reasoning and Planning (CVPR 2024) · better on 5 of 5 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
- Inst3D-LMM: Instance-Aware 3D Scene Understanding with Multi-modal Instruction Tuning (CVPR 2025 · similarity 0.88)
- Merge3D: Efficient 3D Multimodal LLMs via Joint 2D-3D Token Merging (CVPR 2026 · similarity 0.87)
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.86)
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.86)
- On Geometry-Enhanced Parameter-Efficient Fine-Tuning for 3D Scene Segmentation (NeurIPS 2025 · similarity 0.86)
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 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.
