Research map / Image generation / Text-to-image diffusion
Curved Diffusion: A Generative Model With Optical Geometry Control
Andrey Voynov, Amir Hertz, Moab Arar, Shlomi Fruchter, Daniel Cohen-Or
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 per-pixel coordinate conditioning fine-tuning to a base text-to-image model | LDM base model · MS-COCO FID | FID: 17.5 -> 17.5 | No clear effect |
| Add self-attention score reweighting by warped density (ablation) | Curved Diffusion · corner-squeezing warp, 100 prompts x 8 images | saliency center alignment of unwarped images | Improved |
| Condition on per-pixel warp field instead of writing 'fisheye' in the prompt (ablation) | Curved Diffusion · fisheye lens geometry | geometric alignment (qualitative) | Improved |
| Generate with the model instead of post-hoc warping a regular image onto a sphere | Sphere texturing · sphere texture human study | preference (%): 12% (base) -> 58% | Improved |
| Use metric tensor conditioning instead of positional conditioning for sphere textures (ablation) | Curved Diffusion · sphere texture human study | preference (%): 28% -> 40% | Improved |
| Add distance-to-center conditioning to metric-conditioned model (ablation) | Metric-conditioned model · sphere panorama | image quality (qualitative) | Improved |
| Use flipped upside-down coordinate grid | Curved Diffusion · out-of-training-distribution warp | local fidelity (qualitative) | Hurt |
| Apply conditioning to strong fish-eye warp, compare unwarping error with manually distorted base outputs | Curved Diffusion · distortion fidelity, k1 in 10-25 | undistortion displacement error | No clear effect |
Closest papers by meaning
- IllumiCraft: Unified Geometry and Illumination Diffusion for Controllable Video Generation (NeurIPS 2025 · similarity 0.88)
- Diffusion Handles Enabling 3D Edits for Diffusion Models by Lifting Activations to 3D (CVPR 2024 · similarity 0.88)
- Sharp-It: A Multi-view to Multi-view Diffusion Model for 3D Synthesis and Manipulation (CVPR 2025 · similarity 0.87)
- Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models (CVPR 2025 · similarity 0.87)
- Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with Image-Based Surface Representation (ICLR 2025 · similarity 0.87)
- Generative Rendering: Controllable 4D-Guided Video Generation with 2D Diffusion Models (CVPR 2024 · similarity 0.87)
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.
