Research map / Image generation / Image restoration
DiffI2I: Efficient Diffusion Model for Image-to-Image Translation
Bin Xia, Yulun Zhang, Shiyin Wang, Yitong Wang, Xinglong Wu, Yapeng Tian, Wenming Yang, Radu Timotfe, Luc Van Gool
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 diffusion model to estimate IPR (vs no DM, V1) (ablation) | DiffI2I_S2 · CelebA-HQ inpainting | FID: 5.2452 -> 4.7013 | Improved |
| add joint optimization of DM and decoder (vs traditional DM optimization) (ablation) | DiffI2I_S2 · CelebA-HQ inpainting | FID: 5.5221 -> 4.7013 | Improved |
| insert variance noise in reverse DM process (ablation) | DiffI2I_S2 · CelebA-HQ inpainting | FID: 4.7013 -> 4.7796 | Hurt |
| stage-2 DM estimate vs stage-1 with GT input (ablation) | DiffI2I · CelebA-HQ inpainting | FID: 4.4826 -> 4.7013 | No clear effect |
| use L2 loss instead of L_diff (L1) for DM (ablation) | DiffI2I_S2 · CelebA-HQ inpainting | FID: 4.7013 -> 4.7518 | Hurt |
| use KL loss instead of L_diff for DM (ablation) | DiffI2I_S2 · CelebA-HQ inpainting | FID: 4.7013 -> 4.7963 | Hurt |
| concatenate input with GT for CPEN_S1 (vs GT only) (ablation) | DiffI2I_S2 · NTIRE2020 Track1 real-world SR | PSNR: 27.44 -> 27.52 | Improved |
| concatenate input with GT for CPEN_S1 (vs GT only) (ablation) | DiffI2I_S2 · NTIRE2020 Track1 real-world SR | LPIPS: 0.2951 -> 0.2943 | Improved |
| use DiffI2I instead of LDM accelerated with DPM-Solver | real-world SR · NTIRE2020 Track1 real-world SR | PSNR: 24.31 -> 27.52 | Improved |
| use DPM-Solver to accelerate LDM | LDM · NTIRE2020 Track1 real-world SR | PSNR: 25.68 -> 24.31 | Hurt |
| increase DM iterations to 3-4 steps (ablation) | DiffI2I_S2 · inpainting | performance: performance stabilizes beyond 4 iterations | Improved |
| replace DiffIR with DiffI2I (modified DI2Iformer) | inpainting · CelebA-HQ narrow masks | FID: 4.5967 -> 3.8640 | 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.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 1 of 1 shared result
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
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.89)
- Diffusion Models Without Attention (CVPR 2024 · similarity 0.89)
- Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators (CVPR 2025 · similarity 0.89)
- High-Fidelity Diffusion-Based Image Editing (AAAI 2024 · similarity 0.89)
- Relay Diffusion: Unifying diffusion process across resolutions for image synthesis (ICLR 2024 · similarity 0.89)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.88)
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.
