Research map / Image generation / Text-to-image diffusion
SCP-Diff: Spatial-Categorical Joint Prior for Diffusion Based Semantic Image Synthesis
Huan-ang Gao, Mingju Gao, Jiaju Li, Wenyi Li, Rong Zhi, Hao Tang, Hao Zhao
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 joint spatial-categorical inference noise prior instead of standard Gaussian noise (ablation) | ControlNet · Cityscapes | FID: 23.35 -> 10.53 | Improved |
| add joint prior (ablation) | ControlNet · ADE20K | mIoU: 20.73 -> 25.61 | Improved |
| add spatial prior only (ablation) | ControlNet · ADE20K | mIoU: 20.73 -> 20.86 | No clear effect |
| add spatial prior only (ablation) | ControlNet · Cityscapes | FID: 23.35 -> 12.83 | Improved |
| add categorical prior only (ablation) | ControlNet · ADE20K | Acc: 61.14 -> 66.63 | Improved |
| combine spatial and categorical priors into joint prior (ablation) | spatial prior · ADE20K | FID: 16.03 -> 12.66 | Improved |
| apply joint prior to FreestyleNet | FreestyleNet · COCO-Stuff | FID: 14.4 -> 11.3 | Improved |
| add joint prior | ControlNet · ADE20K diversity | LPIPS: 0.59 -> 0.56 | Hurt |
| use categorical prior only (ablation) | ControlNet · Cityscapes | color diversity | Hurt |
| choose denoising start fraction mu in [0.8, 0.9] instead of 1 (ablation) | SCP-Diff · Cityscapes and ADE20K | FID | Improved |
| increase number of reference images used for priors (ablation) | SCP-Diff · ADE20K | FID | Improved |
| select diverse reference images by furthest point sampling in CLIP space (ablation) | SCP-Diff · ADE20K | FID | Improved |
Later papers that compared with this method
1 later accepted paper reports SCP-Diff as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Symmetrical Flow Matching: Unified Image Generation, Segmentation, and Classification with Score-Based Generative Models (AAAI 2026 · better on 1 of 1 shared result)
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.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 5 of 7 shared results
- ECGAN — EC-GAN: Inferring Brain Effective Connectivity via Generative Adversarial Networks (AAAI 2020) · better on 3 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
- BoxDiff: Text-to-Image Synthesis with Training-Free Box-Constrained Diffusion (ICCV 2023 · similarity 0.87)
- ControlNet-XS: Rethinking the Control of Text-to-Image Diffusion Models as Feedback-Control Systems (ECCV 2024 · similarity 0.87)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.87)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.87)
- Stochastic Conditional Diffusion Models for Robust Semantic Image Synthesis (ICML 2024 · similarity 0.87)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · 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.
