Research map / Image generation / Image restoration
SODiff:Semantic-Oriented Diffusion Model for JPEG Compression Artifacts Removal
Tingyu Yang, Jue Gong, Jinpei Guo, Wenbo Li, Yong Guo, Yulun 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 |
|---|---|---|---|
| align image prompt embeddings to text embeddings with an alignment loss (SAIPE vs no alignment) (ablation) | SODiff image prompt extractor · Urban100 | DISTS: 0.1261 -> 0.0862 | Improved |
| align image prompt embeddings to text embeddings with an alignment loss (SAIPE vs no alignment) (ablation) | SODiff image prompt extractor · DIV2K-val | MANIQA: 0.3250 -> 0.4013 | Improved |
| use SAIPE image prompts instead of DAPE text prompts (ablation) | SODiff · Urban100 | MUSIQ: 71.53 -> 72.51 | Improved |
| use SAIPE image prompts instead of DAPE text prompts (ablation) | SODiff · DIV2K-val | DISTS: 0.0697 -> 0.0731 | Hurt |
| add quality-factor-aware timestep predictor (ablation) | SODiff · Ablation (Table 3) | LPIPS: 0.3779 -> 0.3124 | Improved |
| add quality-factor-aware timestep predictor (ablation) | SODiff · Ablation (Table 3) | CLIPIQA: 0.4105 -> 0.5181 | Improved |
| add implicit QF learning loss Lqf (ablation) | SODiff · Ablation (Table 3) | DISTS: 0.1611 -> 0.1469 | Improved |
| add implicit QF learning loss Lqf (ablation) | SODiff · Ablation (Table 3) | MUSIQ: 58.7188 -> 60.3954 | Improved |
| exclude edge-aware DISTS loss LEA (ablation) | SODiff · Ablation (Table 4) | all metrics | Hurt |
| omit adversarial loss LG (ablation) | SODiff · Ablation (Table 4) | perceptual quality vs other criteria | Mixed |
Closest papers by meaning
- Compression-Aware One-Step Diffusion Model for JPEG Artifact Removal (ICCV 2025 · similarity 0.91)
- JPEG Artifacts Removal via Contrastive Representation Learning (ECCV 2022 · similarity 0.88)
- Learning Dual Priors for JPEG Compression Artifacts Removal (ICCV 2021 · similarity 0.88)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · similarity 0.85)
- OSCAR: One-Step Diffusion Codec Across Multiple Bit-rates (NeurIPS 2025 · similarity 0.85)
- ResDiff: Combining CNN and Diffusion Model for Image Super-resolution (AAAI 2024 · similarity 0.85)
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.
