Research map / Image generation / Text-to-image diffusion
RelaxFlow: Text-Driven Amodal 3D Generation
Jiayin Zhu, Guoji Fu, Xiaolu Liu, Qiyuan He, Yicong Li, Angela Yao
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 RelaxFlow dual-branch guidance to SAM3D | SAM3D · ExtremeOcc-3D | CLIP text score / Point-FID: CLIPtxt 24.08 -> 27.26; Point-FID 100.38 -> 81.11 | Improved |
| add RelaxFlow to TRELLIS | TRELLIS · ExtremeOcc-3D | Point-FID: 141.48 -> 97.79 | Improved |
| add low-pass relaxation of prior-branch cross-attention logits (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | Point-FID: 87.1 -> 81.1 | Improved |
| add visibility-aware mask in fusion (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | Point-FID: 92.3 -> 81.1 | Improved |
| raise relaxation strength sigma from 1.0 to 2.5 (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | Point-FID: 81.1 -> 95.2 | Hurt |
| raise prior cutoff rho from 0.2 to 1.0 (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | Point-FID: 81.1 -> 89.9 | Hurt |
| raise prior cutoff rho from 0.2 to 0.4 (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | Point-FID: 81.1 -> 86.5 | Hurt |
| use generated priors instead of retrieved priors (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | Point-FID: 81.1 -> 82.7 | No clear effect |
| increase number of prior images N (ablation) | RelaxFlow (SAM3D) · ExtremeOcc-3D | CLIP text / CLIP image | Mixed |
| use RelaxFlow instead of SAM3D | SAM3D · AmbiSem-3D user study (n=32) | overall preference: 9.22% -> 68.52% | Improved |
| provide prior image as second view to TRELLIS (multi-view) or 2D-edit-then-3D | TRELLIS / SAM3D · AmbiSem-3D | CLIP image score / fidelity | Hurt |
Closest papers by meaning
- Learning Continuous 3D Words for Text-to-Image Generation (CVPR 2024 · similarity 0.87)
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · similarity 0.87)
- Sculpt3D: Multi-View Consistent Text-to-3D Generation with Sparse 3D Prior (CVPR 2024 · similarity 0.87)
- VP3D: Unleashing 2D Visual Prompt for Text-to-3D Generation (CVPR 2024 · similarity 0.86)
- DreamControl: Control-Based Text-to-3D Generation with 3D Self-Prior (CVPR 2024 · similarity 0.86)
- ViewDiff: 3D-Consistent Image Generation with Text-to-Image 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.
