Research map / 3D vision / 3D generation
Aligning Text, Images and 3D Structure Token-by-Token
Aadarsh Sahoo, Vansh Tibrewal, Georgia Gkioxari
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 auxiliary pixel-space reconstruction loss (multi-view) (ablation) | 3D VQ-VAE · Objaverse 3D VQ-VAE (human eval) | Mean rank: 2.828 -> 1.500 | Improved |
| use multiple views instead of a fixed single view for the auxiliary loss (ablation) | 3D VQ-VAE · Objaverse 3D VQ-VAE (human eval) | Mean rank: 1.672 -> 1.500 | Improved |
| use Trellis-based VQ-VAE instead of SAR3D tokenizer | 3D tokenizer · Objaverse 3D tokenizer (human eval) | Mean rank: 1.605 -> 1.395 | Improved |
| coordinate granularity 0.05 instead of 0.5 (ablation) | Kyvo · CLEVR | Recognition Jaccard: 0.2352 -> 0.9212 | Improved |
| coordinate granularity 0.05 instead of 0.005 (ablation) | Kyvo · CLEVR | Recognition Jaccard: 0.5707 -> 0.9212 | Improved |
| coordinate granularity 0.05 instead of 0.005 (ablation) | Kyvo · CLEVR | QA accuracy: 0.5185 -> 0.4980 | Hurt |
| place image before 3D in sequence (ablation) | Kyvo · CLEVR instruction-following | Jaccard: 0.8350 -> 0.8666 | Improved |
| place image before 3D in sequence (ablation) | Kyvo · CLEVR QA | accuracy: 0.4720 -> 0.4980 | Improved |
| add center-token reordering and weighted loss together (ablation) | Kyvo · CLEVR rendering | Rendering mean rank: 2.66 -> 1.00 | Improved |
| add center-token reordering alone (ablation) | Kyvo · CLEVR rendering | Rendering mean rank: 2.66 -> 3.56 | Hurt |
| add weighted loss alone (ablation) | Kyvo · CLEVR rendering | Rendering mean rank: 2.66 -> 2.78 | No clear effect |
| full fine-tuning instead of LoRA (ablation) | Llama-3.2-1B-Instruct · CLEVR | Recognition Jaccard: 0.8684 -> 0.9212 | Improved |
Closest papers by meaning
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.89)
- A3D: Does Diffusion Dream about 3D Alignment? (ICLR 2025 · similarity 0.89)
- SAR3D: Autoregressive 3D Object Generation and Understanding via Multi-scale 3D VQVAE (CVPR 2025 · similarity 0.88)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.88)
- Part-X-MLLM: Part-aware 3D Multimodal Large Language Model (ICLR 2026 · similarity 0.88)
- CG-MLLM: Captioning and Generating 3D content via Multi-modal Large Language Models (ICML 2026 · 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.
