Research map / 3D vision / 3D generation
3DrawAgent: Teaching LLM to Draw in 3D with Early Contrastive Experience
Hongcan Xiao, Xinyue Xiao, Yilin Wang, Yue Zhang, Yonggang Qi
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 contrastive knowledge extraction (CKE) experience library (ablation) | 3DrawAgent base LLM · ModelNet40 test | CLIP-S: 0.5735 -> 0.6643 | Improved |
| use group size K=5 instead of K=2 (ablation) | 3DrawAgent · ModelNet40 test | CLIP-S (Ep 2): 0.6493 -> 0.6643 | Improved |
| use group size K=5 instead of K=10 (ablation) | 3DrawAgent · ModelNet40 test | CLIP-S (Ep 2): 0.5612 -> 0.6643 | Improved |
| provide ground truth during critique (GT=True) (ablation) | 3DrawAgent · ModelNet40 test | CLIP-S (peak): 0.6643 -> 0.6648 | No clear effect |
| provide ground truth during critique (GT=True) (ablation) | 3DrawAgent · ModelNet40 test | CLIP-S (Ep 3): 0.6428 -> 0.6141 | Hurt |
| use CLIP-guided pair selection instead of random pairs (ablation) | 3DrawAgent CKE · ModelNet40 test | CLIP-S (Ep 2): 0.5595 -> 0.6643 | Improved |
| run more CKE epochs (Ep 2 to Ep 3) | 3DrawAgent · ModelNet40 test | CLIP-S: 0.6643 -> 0.6428 | Hurt |
| use Gemini-2.5 Pro instead of DeepSeek-V3.2 as backbone | 3DrawAgent · Text-to-3D category | CLIP-S_T: 0.643 -> 0.649 | Improved |
| use training-free 3DrawAgent vs trained Diff3DS | Diff3DS · Text-to-3D category | AES: 3.791 -> 4.161 | Improved |
| use 3DrawAgent instead of Dream3DVG | Dream3DVG · Text-to-3D category | CLIP-S_T: 0.660 -> 0.649 | No clear effect |
| use 3DrawAgent instead of Diff3DS in user study | Diff3DS · user preference | preference rate: 16.67% -> 46.66% | Improved |
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.
- Diff3DS — Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve Rendering (ICLR 2025) · better on 6 of 6 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
- SketchAgent: Language-Driven Sequential Sketch Generation (CVPR 2025 · similarity 0.88)
- ShapeCraft: LLM Agents for Structured, Textured and Interactive 3D Modeling (NeurIPS 2025 · similarity 0.86)
- SketchThinker-R1: Towards Efficient Sketch-Style Reasoning in Large Multimodal Models (ICLR 2026 · similarity 0.86)
- Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve Rendering (ICLR 2025 · similarity 0.86)
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · similarity 0.86)
- Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models (NeurIPS 2024 · similarity 0.86)
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.
