Research map / Image generation / Text-to-image diffusion
Towards Spatially Consistent Image Generation: On Incorporating Intrinsic Scene Properties into Diffusion Models
Hyundo Lee, Suhyung Choi, Inwoo Hwang, Byoung-Tak Zhang
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 |
|---|---|---|---|
| co-generate intrinsics with I-LDM (Gaussian) instead of base SD2.1 | SD2.1 · Parti prompts | ImageReward: 0.2998 -> 0.4416 | Improved |
| co-generate intrinsics (Gaussian) | SD2.1 · Multi-object prompts | ImageReward: -0.7389 -> -0.5739 | Improved |
| fine-tune SD2.1 with extra CNN channels for intrinsics | SD2.1 · Parti prompts | ImageReward: 0.2998 -> -1.5378 | Hurt |
| double sampling steps of base | SD2.1 · Multi-object prompts | ImageReward: -0.7389 -> -0.7342 | No clear effect |
| remove cross-domain weight scheduling (ablation) | I-LDM · Parti prompts | ImageReward: 0.3852 -> 0.4803 | Improved |
| add intrinsic VAE (ablation) | I-LDM · Parti prompts | ImageReward: 0.3545 -> 0.4803 | Improved |
| add depth intrinsic (ablation) | I-LDM · Parti prompts | ImageReward: 0.3914 -> 0.4803 | Improved |
| add line drawing intrinsic (ablation) | I-LDM · Parti prompts | ImageReward: 0.5074 -> 0.4803 | Hurt |
| add Gaussian-scheduled intrinsics for hand generation | SD2.1 hand LoRA · InterHand | detection rate: 29.82% -> 39.40% | Improved |
| downscale intrinsic resolution during training (ablation) | I-LDM · Parti prompts | ImageReward: 0.4803 (512) vs 0.4960 (256) | No clear effect |
Closest papers by meaning
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.89)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.88)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.88)
- Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis (CVPR 2025 · similarity 0.88)
- RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Intrinsic Image Diffusion for Indoor Single-view Material Estimation (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.
