Research map / Label-efficient and robust learning / Domain adaptation and generalization
Rethinking Knowledge Transfer in Image Quality Assessment: A Perceptual Preference Structure Alignment Perspective
Aobo Li, Jinjian Wu, Yongxu Liu, Jupo Ma, Weisheng Dong
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 |
|---|---|---|---|
| select 20% preference-aligned source samples (PreSTA-S) instead of full KADID-10k | Swin-B trained on KADID-10k · LIVEC cross-domain | SRCC: 0.589 -> 0.744 | Improved |
| select 20% preference-aligned samples of KADID-10k | Swin-B · BID cross-domain | SRCC: 0.739 -> 0.833 | Improved |
| select 20% preference-aligned samples of KonIQ-10k | Swin-B · TID2013 cross-domain | SRCC: 0.452 -> 0.516 | Improved |
| PreSTA-S on KonIQ-10k source for CSIQ target | Swin-B · CSIQ cross-domain | SRCC: 0.723 -> 0.720 | No clear effect |
| PreSTA-S within synthetic: KADID-10k to LIVE with 20% data | Swin-B · LIVE within-domain | SRCC: 0.934 -> 0.933 | No clear effect |
| PreSTA-S within synthetic: KADID-10k to CSIQ with 27.9% data | Swin-B · CSIQ within-domain | SRCC: 0.773 -> 0.840 | Improved |
| add all KADID-10k (cross-domain) to LIVEC training | target-only Swin-B · LIVEC joint training | SRCC: 0.883 -> 0.876 | Hurt |
| add all KonIQ-10k + SPAQ (21,198 samples) to LIVEC training | target-only Swin-B · LIVEC joint training | SRCC: 0.883 -> 0.899 | Improved |
| add only 2,793 preference-aligned samples (PreSTA-J) instead of all 21,198 | joint training on LIVEC · LIVEC joint training | SRCC: 0.899 -> 0.905 | Improved |
| add 5,343 preference-aligned samples (PreSTA-J) instead of all 21,198 | joint training on BID · BID joint training | SRCC: 0.883 -> 0.885 | Improved |
| use PPC to choose between KADID-10k and KonIQ-10k as source | transfer source selection · 12 backbone-target comparisons | SRCC: higher PPC source wins in 11 of 12 | Improved |
Closest papers by meaning
- Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare (NeurIPS 2024 · similarity 0.89)
- Refine-IQA: Multi-Stage Reinforcement Finetuning for Perceptual Image Quality Assessment (AAAI 2026 · similarity 0.88)
- IQA-Spider: Unifying Multi-Granularity Image Quality Assessment with Reasoning, Grounding and Referring (ICML 2026 · similarity 0.87)
- LGDM: Latent Guidance in Diffusion Models for Perceptual Evaluations (ICML 2025 · similarity 0.87)
- Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Ranking (ICLR 2026 · similarity 0.87)
- Toward Generalized Image Quality Assessment: Relaxing the Perfect Reference Quality Assumption (CVPR 2025 · 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.
