Research map / Point clouds / Point cloud classification
ShapeLLM: Universal 3D Object Understanding for Embodied Interaction
Zekun Qi, Runpei Dong, Shaochen Zhang, Haoran Geng, Chunrui Han, Zheng Ge, Li Yi, Kaisheng Ma
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 multi-view alignment on top of scaled-up ReCon++ (ablation) | ReCon++ · Objaverse-LVIS zero-shot | Top-1 accuracy: 51.5 -> 53.7 | Improved |
| scale up model and pretraining data of ReCon (ablation) | ReCon++ · Objaverse-LVIS zero-shot | Top-1 accuracy: 10.3 -> 53.7 | Improved |
| scale up model and pretraining data of ReCon (ablation) | ShapeLLM · 3D MM-Vet | Total: 21.9 -> 53.1 | Improved |
| add multi-view alignment on top of scaled-up ReCon++ (ablation) | ShapeLLM · 3D MM-Vet | Total: 48.2 -> 53.1 | Improved |
| replace ULIP-2 encoder in PointLLM-13B with ReCon++ (ablation) | PointLLM-13B · 3D MM-Vet | Total: +4.20% | Improved |
| train PointLLM on ShapeLLM's GPT-4V instruction data | PointLLM-13B · 3D MM-Vet | Total | Improved |
| use 3D point cloud inputs instead of 2D images for 6-DoF grounding | ShapeLLM-13B · GAPartNet referring grounding | Avg accuracy: 6.2 -> 11.7 | Improved |
| add absolute position encoding and prompts on top of global representation only (ablation) | ShapeLLM-13B · 3D MM-Vet | Total: 52.4 -> 53.1 | Improved |
| use causal masking instead of random masking in ReCon++ pretraining (ablation) | ReCon++-L · ScanObjectNN fine-tune vs zero-shot | accuracy: fine-tune 92.5 -> 95.3; zero-shot 52.8 -> 49.8 | Mixed |
| backpropagate global-branch gradients to local branch (no stop-gradient) (ablation) | ReCon++-L, random mask · Objaverse-LVIS zero-shot | Top-1 accuracy: 52.8 -> 53.7 | Improved |
| use ShapeLLM instead of language-only LLMs on 3D MM-Vet | ShapeLLM · 3D MM-Vet | Total: 16.2 -> 39.0 (ShapeLLM-7B in language-only table) | Improved |
Later papers that compared with this method
2 later accepted papers report ShapeLLM-13B as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- 4DPC$^2$hat: Towards Dynamic Point Cloud Understanding with Failure-Aware Bootstrapping (ICML 2026 · better on 5 of 5 shared results)
- Exploring the Potential of Encoder-free Architectures in 3D LMMs (ICLR 2026 · better on 2 of 2 shared results)
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-13B — PointLLM: Empowering Large Language Models to Understand Point Clouds (ECCV 2024) · better on 3 of 3 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
- LL3DA: Visual Interactive Instruction Tuning for Omni-3D Understanding Reasoning and Planning (CVPR 2024 · similarity 0.89)
- SpatialLLM: A Compound 3D-Informed Design towards Spatially-Intelligent Large Multimodal Models (CVPR 2025 · similarity 0.89)
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.88)
- ShapeLLM-Omni: A Native Multimodal LLM for 3D Generation and Understanding (NeurIPS 2025 · similarity 0.88)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.88)
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · 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.
