Research map / Image generation / Super-resolution
Comparison of Full-Reference Image Quality Models for Optimization of Image Processing Systems
Keyan Ding, Kede Ma, Shiqi Wang, Eero P. Simoncelli
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 DISTS or LPIPS instead of MAE as loss | DNN denoiser · Denoising | SRCC vs subjective ranking: DISTS 0.346 vs MAE 0.527; LPIPS 0.709 | Mixed |
| use LPIPS instead of MAE as loss | DNN deblurrer · Deblurring | SRCC vs subjective ranking: 0.164 -> 0.855 | Improved |
| use DISTS instead of MAE as loss | DNN super-resolution · Super-resolution | SRCC vs subjective ranking: 0.309 -> 0.782 | Improved |
| use DISTS instead of MAE as loss | DNN compression · Compression | SRCC vs subjective ranking: 0.455 -> 0.855 | Improved |
| replace MAE with MS-SSIM as loss | DNN image restoration (4 tasks) · Denoising, deblurring, super-resolution, compression | human ranking | No clear effect |
| replace LPIPS with DISTS as loss | DNN image restoration · Super-resolution | human ranking | Improved |
| replace VGG-based LPIPS loss with DISTS loss and fine-tune DeblurGAN-v2 | DeblurGAN-v2 · Köhler deblurring | DISTS value: 0.22 -> 0.18 | Improved |
| replace original DeblurGAN-v2 loss with DISTS plus 0.001 adversarial loss | DeblurGAN-v2 · Köhler deblurring | visual quality | Improved |
| pretrain network with MAE before IQA fine-tuning | IQA-optimized DNNs · reference recovery and 4 tasks | optimization stability | Improved |
| initialize recovery from JPEG image instead of white noise (ablation) | surjective IQA models (GTI-CNN, DeepIQA, PieAPP) · Reference image recovery | recovered structure | Improved |
| use surjective mapping IQA models as objective | GTI-CNN, DeepIQA, PieAPP · Reference image recovery | image quality of result | Hurt |
| use injective mapping IQA models as objective | MS-SSIM, PAMSE, NLPD, DISTS · Reference image recovery | recovery of reference | Improved |
Closest papers by meaning
- Toward Generalized Image Quality Assessment: Relaxing the Perfect Reference Quality Assumption (CVPR 2025 · similarity 0.90)
- Augmenting Perceptual Super-Resolution via Image Quality Predictors (CVPR 2025 · similarity 0.90)
- Boosting Image Quality Assessment through Efficient Transformer Adaptation with Local Feature Enhancement (CVPR 2024 · similarity 0.89)
- Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare (NeurIPS 2024 · similarity 0.89)
- Causal-IQA: Towards the Generalization of Image Quality Assessment Based on Causal Inference (ICML 2024 · similarity 0.88)
- Towards Syn-to-Real IQA: A Novel Perspective on Reshaping Synthetic Data Distributions (NeurIPS 2025 · 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.
