Research map / 3D vision / Face and shape
Beyond Prompts: Unconditional 3D Inversion for Out-of-Distribution Shapes
Victoria Chen, Emery Pierson, Léopold Maillard, Maks Ovsjanikov
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 |
|---|---|---|---|
| use an empty prompt instead of an approximate prompt in Euler inversion (ablation) | TRELLIS Euler inversion · DT4D inversion reconstruction | L1: 72.06 -> 1.99 | Improved |
| use an empty prompt instead of an approximate prompt in Euler inversion (ablation) | TRELLIS Euler inversion · DT4D inversion reconstruction | LPIPS: 0.34 -> 0.25 | Improved |
| use an empty prompt instead of an approximate prompt in NTI (ablation) | TRELLIS NTI · DT4D inversion reconstruction | L1: 15.57 -> 1.55 | Improved |
| use an empty prompt instead of an approximate prompt in NTI (ablation) | TRELLIS NTI · DT4D inversion reconstruction | LPIPS: 0.28 -> 0.24 | Improved |
| add null-text optimization to Euler inversion with approximate prompt (ablation) | TRELLIS Euler + approximate prompt · DT4D inversion reconstruction | L1: 72.06 -> 15.57 | Improved |
| add null-text optimization to Euler inversion with empty prompt (ablation) | TRELLIS Euler + empty prompt · DT4D inversion reconstruction | L1: 1.99 -> 1.55 | Improved |
| add null-text optimization to Euler inversion with empty prompt (ablation) | TRELLIS Euler + empty prompt · DT4D inversion reconstruction | LPIPS: 0.25 -> 0.24 | No clear effect |
| use an empty prompt instead of the ground-truth caption for inversion (ablation) | Stable Diffusion v1.4 · SD v1.4 inversion (COCO) | PSNR: 15.73 -> 22.79 | Improved |
| use an empty prompt instead of the ground-truth caption for inversion (ablation) | Stable Diffusion v1.4 · SD v1.4 inversion (COCO) | LPIPS: 0.595 -> 0.507 | Improved |
| use an empty prompt instead of the ground-truth caption for inversion (ablation) | Stable Diffusion v1.4 · SD v1.4 inversion (DiffusionDB) | PSNR: 19.19 -> 25.71 | Improved |
| use an empty prompt instead of the ground-truth caption for inversion (ablation) | Stable Diffusion v1.4 · SD v1.4 inversion (DiffusionDB) | LPIPS: 0.448 -> 0.388 | Improved |
| vary the text prompt semantically across a subject class (text diversity vs visual diversity) | TRELLIS text-to-3D · Prompt diversity (8 subjects) | diversity ratio R: R 0.515 (Surgeon) to 1.205 (Dancing girl) | Mixed |
Closest papers by meaning
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · similarity 0.88)
- TextField3D: Towards Enhancing Open-Vocabulary 3D Generation with Noisy Text Fields (ICLR 2024 · similarity 0.87)
- Let 2D Diffusion Model Know 3D-Consistency for Robust Text-to-3D Generation (ICLR 2024 · similarity 0.87)
- ShapeWords: Guiding Text-to-Image Synthesis with 3D Shape-Aware Prompts (CVPR 2025 · similarity 0.87)
- VP3D: Unleashing 2D Visual Prompt for Text-to-3D Generation (CVPR 2024 · similarity 0.87)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 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.
