Research map / 3D vision / 3D generation
SemGrasp: Semantic Grasp Generation via Language Aligned Discretization
Kailin Li, Jingbo Wang, Lixin Yang, Cewu Lu, Bo Dai
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 the refinement token to the grasp representation (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | MPVPE (mm): 20.36 -> 14.97 | Improved |
| use three hierarchical tokens instead of a single token (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | MPVPE (mm): 29.95 -> 14.97 | Improved |
| use more than one refinement token (2 or 3) (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | SIV (cm3): 2.72 -> 2.98 | Hurt |
| use hierarchical VQ-VAE instead of a single shared VQ-VAE (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | MPVPE (mm): 28.02 -> 14.97 | Improved |
| assign semantic meaning (separate losses) to the three tokens (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | MPVPE (mm): 21.94 -> 14.97 | Improved |
| use 512 codebook entries instead of 256 or 1024 (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | MPVPE (mm): 95.52 -> 14.97 | Improved |
| add EMA and codebook reset to VQ-VAE training (ablation) | SemGrasp VQ-VAE · OakInk reconstruction | MPVPE (mm): 14.97 -> 24.67 | Hurt |
| use Vicuna-7B instead of Llama-7B (ablation) | SemGrasp MLLM · language-guided grasp generation | GPT-4 score: 58.9 -> 74.5 | Improved |
| add object size token (ablation) | SemGrasp MLLM · language-guided grasp generation | GPT-4 score: 43.3 -> 74.5 | Improved |
| use two-stage training (alignment then instruction tuning) (ablation) | SemGrasp MLLM · language-guided grasp generation | GPT-4 score: 62.5 -> 74.5 | Improved |
| increase LoRA rank from 64 to 128 (ablation) | SemGrasp MLLM · language-guided grasp generation | GPT-4 score: 74.5 -> 74.5 | No clear effect |
| use MLLM instead of BERT classification baseline | language-guided grasp generation · OakInk test | GPT-4 score: 47.3 -> 74.5 | Improved |
Closest papers by meaning
- Grasp as You Say: Language-guided Dexterous Grasp Generation (NeurIPS 2024 · similarity 0.89)
- DextER: Language-driven Dexterous Grasp Generation with Embodied Reasoning (CVPR 2026 · similarity 0.89)
- Language-driven Grasp Detection (CVPR 2024 · similarity 0.89)
- RAGG: Retrieval-Augmented Grasp Generation Model (AAAI 2025 · similarity 0.89)
- Single-View Scene Point Cloud Human Grasp Generation (CVPR 2024 · similarity 0.87)
- Grasp2Grasp: Vision-Based Dexterous Grasp Translation via Schrödinger Bridges (NeurIPS 2025 · similarity 0.87)
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.
