Research map / Image generation / Autoregressive and latent generation
RankIQA: Learning From Rankings for No-Reference Image Quality Assessment
Xialei Liu, Joost van de Weijer, Andrew D. Bagdanov
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 ranking pre-training on synthetic distortions followed by fine-tuning on TID2013 (ablation) | VGG-16 NR-IQA · TID2013 (all distortions, 80/20 split) | SROCC: 0.612 -> 0.780 | Improved |
| add fine-tuning on TID2013 after ranking pre-training (ablation) | RankIQA · TID2013 (all distortions, 80/20 split) | SROCC: 0.623 -> 0.780 | Improved |
| train on generated rankings only, without any TID2013 data (ablation) | VGG-16 · TID2013 (all distortions, 80/20 split) | SROCC: 0.612 -> 0.623 | Improved |
| use ranking-only network instead of direct fine-tuning on a hard distortion type (#13) (ablation) | RankIQA · TID2013 distortion #13 | SROCC: 0.725 -> 0.431 | Hurt |
| use a deeper network (AlexNet instead of Shallow) for ranking training (ablation) | RankIQA · LIVE (from scratch, 80/20 reference-image split) | SROCC: 0.937 -> 0.949 | Improved |
| use VGG-16 instead of AlexNet for ranking training (ablation) | RankIQA · LIVE (from scratch, 80/20 reference-image split) | SROCC: 0.949 -> 0.974 | Improved |
| initialize ranking network from ImageNet pre-trained weights (ablation) | RankIQA VGG-16 · LIVE | LCC and SROCC | Improved |
| use efficient Siamese backpropagation over all batch pairs instead of random pairs or hard-negative mining (ablation) | Siamese ranking network (AlexNet) · JPEG distortion ranking loss | convergence of ranking loss | Improved |
| generate ranked images from Places2 instead of Waterloo (ablation) | RankIQA+FT · LIVE (ALL) | SROCC: 0.981 -> 0.980 | No clear effect |
| generate ranked images from Places2 instead of Waterloo (ablation) | RankIQA+FT · LIVE (ALL) | LCC: 0.982 -> 0.981 | No clear effect |
| use RankIQA+FT instead of the best prior NR-IQA (HOSA) | NR-IQA on TID2013 · TID2013 (all distortions, 80/20 split) | SROCC: 0.728 -> 0.780 | Improved |
Closest papers by meaning
- Adaptive Feature Selection for No-Reference Image Quality Assessment by Mitigating Semantic Noise Sensitivity (ICML 2024 · similarity 0.88)
- Transformer-Based No-Reference Image Quality Assessment via Supervised Contrastive Learning (AAAI 2024 · similarity 0.88)
- Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare (NeurIPS 2024 · similarity 0.87)
- Boosting Image Quality Assessment through Efficient Transformer Adaptation with Local Feature Enhancement (CVPR 2024 · similarity 0.87)
- VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank (NeurIPS 2025 · similarity 0.87)
- IQA-Spider: Unifying Multi-Granularity Image Quality Assessment with Reasoning, Grounding and Referring (ICML 2026 · similarity 0.87)
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.
