Research map / 3D vision / 3D generation
LATTE3D: Large-scale Amortized Text-To-Enhanced3D Synthesis
Kevin Xie, Tianshi Cao, Jonathan P Lorraine, Jun Gao, James R Lucas, Antonio Torralba, Sanja Fidler, Xiaohui Zeng
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 MVDream 3D-aware diffusion guidance to ATT3D (ablation) | ATT3D · Seen prompts | Render-FID: 275.01 -> 236.73 | Improved |
| replace hypernetwork with LATTE3D triplane U-Net architecture (ablation) | ATT3D+MV · Seen prompts | Render-FID: 236.73 -> 203.33 | Improved |
| add reconstruction pretraining (ablation) | ATT3D+MV+UNet · Seen prompts | Render-FID: 203.33 -> 189.51 | Improved |
| add shape regularization loss (ablation) | ATT3D+MV+UNet+PT · Seen prompts | Mask-FID: 189.09 -> 176.44 | Improved |
| add point cloud annealing during stage-1 training (ablation) | LATTE3D · Unseen DreamFusion prompts, dummy input | FID: 193.28 -> 190.00 | Improved |
| add stage-2 surface-based texture refinement (ablation) | LATTE3D · qualitative comparison | texture quality | Improved |
| add 10 minutes of test-time optimization | LATTE3D · Unseen DreamFusion prompts | Render-FID: 190.00 -> 178.37 | Improved |
| add test-time optimization | LATTE3D · Unseen DreamFusion prompts | CLIP-Score: 0.2605 -> 0.3000 | Improved |
| amortize over many prompts instead of per-prompt MVDream optimization | LATTE3D · Optimization cost for matched user preference | GPU time per prompt: 36 GPU minutes -> 215 GPU seconds | Improved |
| train both geometry and texture in stage-2 instead of freezing geometry (ablation) | LATTE3D · stage-2 training | stability | Hurt |
Closest papers by meaning
- Instant3D: Fast Text-to-3D with Sparse-view Generation and Large Reconstruction Model (ICLR 2024 · similarity 0.88)
- HIFA: High-fidelity Text-to-3D Generation with Advanced Diffusion Guidance (ICLR 2024 · similarity 0.88)
- DIRECT-3D: Learning Direct Text-to-3D Generation on Massive Noisy 3D Data (CVPR 2024 · similarity 0.87)
- One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View Generation and 3D Diffusion (CVPR 2024 · similarity 0.87)
- ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models (CVPR 2024 · similarity 0.87)
- Structured 3D Latents for Scalable and Versatile 3D Generation (CVPR 2025 · 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.
