Research map / Point clouds / Point cloud classification
Test-Time Optimization of 3D Point Cloud LLM via Manifold-Aware In-Context Guidance and Refinement
Tiankai Chen, Nanqing Liu, Li Yang, Xulei Yang, Tianrui Li, Xun Xu
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 in-context guidance (graph-based neighbor captions) (ablation) | PGLLM · ModelNet40 OOD | AUROC: 80.4 -> 83.1 | Improved |
| add in-context guidance (ablation) | PGLLM · ModelNet40 recognition | ACC: 52.5 -> 60.2 | Improved |
| add score propagation alone (ablation) | PGLLM · ModelNet40 OOD | FPR95: 100.0 -> 62.0 | Improved |
| add score propagation alone (ablation) | PGLLM · ModelNet40 recognition | ACC: 52.5 -> 56.7 | Improved |
| combine in-context guidance with score propagation (ablation) | PGLLM · ModelNet40 OOD | AUROC: 83.1 -> 85.9 | Improved |
| combine in-context guidance with score propagation (ablation) | PGLLM · ModelNet40 OOD | FPR95: 100.0 -> 52.1 | Improved |
| replace graph with direct nearest-sample retrieval for guidance (ablation) | PGLLM · ModelNet40 recognition | ACC: 59.7 -> 63.1 | No clear effect |
| use GPT-4 instead of DeepSeek-V3 as second-stage LLM | PGLLM_T · ModelNet40 OOD | AUROC: +3.8% | Improved |
| use Llama3.1-8B instead of Qwen3-VL-8B as second-stage LLM | PGLLM_T (PointLLM-7B) · ModelNet40 OOD | AUROC: 81.5 -> 69.6 | Hurt |
| use Qwen-Plus as second-stage LLM for recognition | PGLLM_T · ModelNet40 recognition | ACC: 62.5 -> 42.4 | Hurt |
| incrementally add test samples to graph instead of full graph (ablation) | PGLLM_O/T · ModelNet40 OOD | AUROC: 89.6 -> 86.8 | Hurt |
| use PGLLM over MiniGPT-3D baseline | PGLLM_T · ModelNet40 recognition | ACC: 60.9 -> 62.5 | 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.
- PointLLM-7B — PointLLM: Empowering Large Language Models to Understand Point Clouds (ECCV 2024) · 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
- More Text, Less Point: Towards 3D Data-Efficient Point-Language Understanding (AAAI 2025 · similarity 0.91)
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.90)
- Language-Image Models with 3D Understanding (ICLR 2025 · similarity 0.89)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.89)
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · similarity 0.89)
- ObjVariantEnsemble: Advancing Point Cloud LLM Evaluation in Challenging Scenes with Subtly Distinguished Objects (AAAI 2025 · similarity 0.89)
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.
