AI Paper Atlas

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

ICLR 2026 · Point clouds › Point cloud classification · approach: Point-language and open-vocabulary 3D

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 changedWhereMeasuredEffect
add in-context guidance (graph-based neighbor captions) (ablation)PGLLM · ModelNet40 OODAUROC: 80.4 -> 83.1Improved
add in-context guidance (ablation)PGLLM · ModelNet40 recognitionACC: 52.5 -> 60.2Improved
add score propagation alone (ablation)PGLLM · ModelNet40 OODFPR95: 100.0 -> 62.0Improved
add score propagation alone (ablation)PGLLM · ModelNet40 recognitionACC: 52.5 -> 56.7Improved
combine in-context guidance with score propagation (ablation)PGLLM · ModelNet40 OODAUROC: 83.1 -> 85.9Improved
combine in-context guidance with score propagation (ablation)PGLLM · ModelNet40 OODFPR95: 100.0 -> 52.1Improved
replace graph with direct nearest-sample retrieval for guidance (ablation)PGLLM · ModelNet40 recognitionACC: 59.7 -> 63.1No clear effect
use GPT-4 instead of DeepSeek-V3 as second-stage LLMPGLLM_T · ModelNet40 OODAUROC: +3.8%Improved
use Llama3.1-8B instead of Qwen3-VL-8B as second-stage LLMPGLLM_T (PointLLM-7B) · ModelNet40 OODAUROC: 81.5 -> 69.6Hurt
use Qwen-Plus as second-stage LLM for recognitionPGLLM_T · ModelNet40 recognitionACC: 62.5 -> 42.4Hurt
incrementally add test samples to graph instead of full graph (ablation)PGLLM_O/T · ModelNet40 OODAUROC: 89.6 -> 86.8Hurt
use PGLLM over MiniGPT-3D baselinePGLLM_T · ModelNet40 recognitionACC: 60.9 -> 62.5Improved

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy