Research map / Image generation / Autoregressive and latent generation
Improving Diffusion-Based Image Synthesis with Context Prediction
Ling Yang, Jingwei Liu, Shenda Hong, Zhilong Zhang, Zhilin Huang, Zheming Cai, Wentao Zhang, Bin CUI
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 neighborhood context prediction | Continuous diffusion backbone · MS-COCO 256×256 zero-shot | FID-30K: 7.27 -> 6.21 | Improved |
| Add neighborhood context prediction | Discrete diffusion backbone · MS-COCO 256×256 zero-shot | FID-30K: 8.44 -> 6.67 | Improved |
| Add neighborhood context prediction | Diffusion inpainting model · CelebA-HQ Wide | LPIPS: 0.045 -> 0.042 | Improved |
| Add neighborhood context prediction | Diffusion inpainting model · CelebA-HQ Altern. Lines | LPIPS: 0.009 -> 0.022 | Hurt |
| Add neighborhood context prediction | Diffusion inpainting model · ImageNet Wide | LPIPS: 0.105 -> 0.098 | Improved |
| Add neighborhood context prediction | Diffusion inpainting model · ImageNet Altern. Lines | LPIPS: 0.089 -> 0.107 | Hurt |
| Add neighborhood context prediction | LDM diffusion backbone · FFHQ 256×256 | FID: 4.98 -> 2.24 | Improved |
| Add neighborhood context prediction | LDM diffusion backbone · LSUN-Bedrooms 256×256 | FID: 2.95 -> 1.12 | Improved |
| Add neighborhood context prediction | LDM diffusion backbone · CelebA-HQ 256×256 | FID: 5.11 -> 3.22 | Improved |
| Add neighborhood context prediction | LDM diffusion backbone · LSUN-Churches 256×256 | FID: 4.02 -> 1.78 | Hurt |
| Increase the number of neighborhood strides (ablation) | ConPreDiff with LDM backbone · Image generation ablation | FID | Improved |
| Use 1-stride neighbors in context prediction (ablation) | ConPreDiff with LDM backbone · Image generation ablation | FID | Improved |
Closest papers by meaning
- MC-DiT: Contextual Enhancement via Clean-to-Clean Reconstruction for Masked Diffusion Models (NeurIPS 2024 · similarity 0.90)
- DCTdiff: Intriguing Properties of Image Generative Modeling in the DCT Space (ICML 2025 · similarity 0.89)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.89)
- Any-Size-Diffusion: Toward Efficient Text-Driven Synthesis for Any-Size HD Images (AAAI 2024 · similarity 0.89)
- Adapting Diffusion Models for Improved Prompt Compliance and Controllable Image Synthesis (NeurIPS 2024 · similarity 0.89)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.89)
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.
