Research map / Image generation / Image restoration
Efficient Zero-shot Inpainting with Decoupled Diffusion Guidance
Badr MOUFAD, Yazid Janati, Navid Bagheri Shouraki, Alain Oliviero Durmus, Thomas Hirtz, Eric Moulines, Jimmy Olsson
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 |
|---|---|---|---|
| use DING instead of FlowChef | FlowChef · FFHQ 768 (Half) | FID: 20.2 -> 9.6 | Improved |
| use DING instead of DiffPIR | DiffPIR · FFHQ 768 (Half) | cPSNR: 30.91 -> 31.03 | Hurt |
| compare DING to Blended on observed-region fidelity | Blended · FFHQ 768 (Half) | cPSNR: 31.32 -> 31.03 | Hurt |
| use DING instead of DiffPIR | DiffPIR · DIV2K 768 (Half) | FID: 41.1 -> 39.2 | Improved |
| use DING instead of DDNM | DDNM · PIE-Bench | FID: 61.4 -> 61.4 | No clear effect |
| use DING instead of Blended | Blended · PIE-Bench | CLIP-ED: 23.15 -> 23.36 | Improved |
| use DING instead of fine-tuned SD3 inpaint (28 NFEs) | SD3 Inpaint · FFHQ 512 (Half) | cPSNR: 21.69 -> 31.40 | Improved |
| use DING instead of fine-tuned SD3 inpaint | SD3 Inpaint · PIE-Bench | FID: 68.7 -> 63.6 | Improved |
| give fine-tuned SD3 inpaint 56 NFEs instead of 28 (ablation) | SD3 Inpaint · FFHQ 512 (Half) | FID: 23.5 -> 23.7 | No clear effect |
| evaluate denoiser at proxy variable (2 NFEs/step) instead of reusing previous noise prediction (ablation) | Delayed DING · DIV2K 768 (Half) | FID: 43.9 -> 39.2 | Improved |
| evaluate denoiser at proxy variable instead of reusing previous noise prediction (ablation) | Delayed DING · FFHQ 768 (Half) | FID: 9.1 -> 9.6 | Mixed |
| use default fast-decaying eta schedule instead of rescaled DDPM (B) (ablation) | DING · FFHQ 768 (Half) | FID: 21.5 -> 9.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FlowChef — FlowChef: Steering of Rectified Flow Models for Controlled Generations (ICCV 2025) · better on 5 of 5 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
- PostEdit: Posterior Sampling for Efficient Zero-Shot Image Editing (ICLR 2025 · similarity 0.90)
- InvFusion: Bridging Supervised and Zero-shot Diffusion for Inverse Problems (NeurIPS 2025 · similarity 0.90)
- High-Fidelity Diffusion-Based Image Editing (AAAI 2024 · similarity 0.89)
- Reconciling Stochastic and Deterministic Strategies for Zero-shot Image Restoration using Diffusion Model in Dual (CVPR 2025 · similarity 0.89)
- RAD: Region-Aware Diffusion Models for Image Inpainting (CVPR 2025 · similarity 0.89)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (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.
