Research map / 3D vision / Face and shape
SuperVoxelGPT: Adaptive and Ordered 3D Tokenization for Autoregressive Shape Generation
Yuan Li, Congyi Zhang, Xifeng Gao, Xiaohu Guo
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 saliency-guided CVT instead of uniform CVT at same compression (ablation) | SuperVoxelGPT · Image-to-3D | CD_L2: 0.0157 -> 0.0122 | Improved |
| add saliency-guided CVT instead of uniform CVT (ablation) | SuperVoxelGPT · Image-to-3D | PSNR: 29.09 -> 32.08 | Improved |
| add SuperVoxelVAE local encoding instead of CraftsMan3D set-based encoder with FSQ (ablation) | SuperVoxelGPT · Image-to-3D | CD_L2: 0.0402 -> 0.0122 | Improved |
| add SuperVoxelVAE local encoding instead of set-based encoder (ablation) | SuperVoxelGPT · Image-to-3D | PSNR: 20.91 -> 32.08 | Improved |
| add SuperVoxelVAE encoding instead of set-based encoder (ablation) | SuperVoxelGPT · Image-to-3D | Spatial ordering compressibility gap: 0.0060 -> 0.1254 | Improved |
| add SuperVoxelVAE encoding instead of set-based encoder (ablation) | SuperVoxelGPT · Image-to-3D | APCU: 0.5649 -> 0.7727 | Improved |
| use Jacobi decoding instead of token-by-token decoding (ablation) | SuperVoxelGPT · Image-to-3D | MLLM inference time (s): 47.21 -> 1.20 | Improved |
| use Jacobi decoding instead of token-by-token decoding (ablation) | SuperVoxelGPT · Image-to-3D | CD_L2 / PSNR: 0.0122 -> 0.0122; 32.09 -> 32.08 | No clear effect |
| use Jacobi decoding instead of token-by-token decoding (ablation) | SuperVoxelGPT · Image-to-3D | Token accuracy (%): 97.80 -> 97.50 | No clear effect |
| increase K and saliency threshold t (ablation) | Saliency CVT · Parameter sweep (1,000 test shapes) | compression vs reconstruction CD_L2: K=4, t=0.1 gives c=12.8% | Mixed |
| use t=0.1 instead of t=0 (ablation) | Saliency CVT · Parameter sweep | reconstruction CD_L2 | No clear effect |
| use K=4 instead of K=6 or K=8 (ablation) | Saliency CVT · Parameter sweep | salient-region fidelity | Improved |
Closest papers by meaning
- Part-X-MLLM: Part-aware 3D Multimodal Large Language Model (ICLR 2026 · similarity 0.88)
- SAR3D: Autoregressive 3D Object Generation and Understanding via Multi-scale 3D VQVAE (CVPR 2025 · similarity 0.88)
- CG-MLLM: Captioning and Generating 3D content via Multi-modal Large Language Models (ICML 2026 · similarity 0.87)
- ShapeLLM-Omni: A Native Multimodal LLM for 3D Generation and Understanding (NeurIPS 2025 · similarity 0.87)
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.87)
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · 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.
