Research map / 3D vision / 3D generation
3D-LATTE: Latent Space 3D Editing from Textual Instructions
Maria Parelli, Michael Oechsle, Michael Niemeyer, Federico Tombari, Andreas Geiger
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 |
|---|---|---|---|
| edit in latent space of a 3D diffusion model instead of Edit360 | 3D editing · 3D edit benchmark | CLIP directional similarity: 0.149 -> 0.178 | Improved |
| edit in latent space instead of MVEdit | 3D editing · 3D edit benchmark | CLIP direction consistency: 0.67 -> 0.77 | Improved |
| compare with GaussCTRL on shape preservation | 3D-LATTE · 3D edit benchmark | CLIP diff no-edit: 0.035 -> 0.039 | Hurt |
| add 3D enhancement via iterative dataset update (ablation) | 3D-LATTE · qualitative renders | visual detail | Improved |
| add geometry regularization guidance (ablation) | 3D-LATTE · qualitative edits | geometry integrity | Improved |
| add frequency annealing (ablation) | 3D-LATTE · qualitative edits | texture noise | Improved |
| prefer 3D-LATTE over GaussCTRL in GPTEval3D | 3D-LATTE · 3D edit benchmark | text prompt alignment win rate: 94% | Improved |
| prefer 3D-LATTE over Edit360 on plausibility | 3D-LATTE · 3D edit benchmark | 3D plausibility win rate: 90% | Improved |
| user study vs three baselines | 3D-LATTE · user study | instruction faithfulness votes: 83.2% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Edit360 — Edit360: 2D Image Edits to 3D Assets from Any Angle (ICCV 2025) · better on 3 of 3 shared results
- GaussCTRL — GaussCtrl: Multi-View Consistent Text-Driven 3D Gaussian Splatting Editing (ECCV 2024) · better on 2 of 3 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
- SceneFactor: Factored Latent 3D Diffusion for Controllable 3D Scene Generation (CVPR 2025 · similarity 0.89)
- SyncNoise: Geometrically Consistent Noise Prediction for Instruction-based 3D Editing (AAAI 2025 · similarity 0.89)
- EditRoom: LLM-parameterized Graph Diffusion for Composable 3D Room Layout Editing (ICLR 2025 · similarity 0.88)
- EditSplat: Multi-View Fusion and Attention-Guided Optimization for View-Consistent 3D Scene Editing with 3D Gaussian Splatting (CVPR 2025 · similarity 0.88)
- A3D: Does Diffusion Dream about 3D Alignment? (ICLR 2025 · similarity 0.88)
- SHAP-EDITOR: Instruction-Guided Latent 3D Editing in Seconds (CVPR 2024 · 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.
