Research map / Label-efficient and robust learning / Test-time adaptation
Test Time Adaptation for Blind Image Quality Assessment
Subhadeep Roy, Shankhanil Mitra, Soma Biswas, Rajiv Soundararajan
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 TTA-IQA (rank + group contrastive loss) to source model | TReS, MUSIQ, HyperIQA, MetaIQA · PIPAL (large shift) | SROCC: 0.3845 -> 0.4278 (TReS) | Improved |
| add TTA-IQA to source model on LIVE-IQA | TReS · LIVE-IQA | SROCC: 0.5435 -> 0.6722 | Improved |
| use quality-aware rank+GC loss instead of rotation prediction as auxiliary task | TTA-IQA · KonIQ-10k, MUSIQ | SROCC: 0.6577 -> 0.6693 | Improved |
| rotation prediction beats TTA-IQA on HyperIQA for KonIQ-10k and LIVE-IQA | HyperIQA · KonIQ-10k | SROCC: TTA-IQA 0.5960 vs rotation 0.6033 | Mixed |
| add rank loss to GC loss (ablation) | TTA-IQA · PIPAL, TReS | SROCC: 0.4666 -> 0.4278 | Hurt |
| combine rank and GC loss versus either alone (ablation) | TTA-IQA · MUSIQ and MetaIQA across datasets | SROCC: MetaIQA KonIQ 0.5414/0.5580 -> 0.5838 | Improved |
| GC loss on extremely distorted images versus rank loss (ablation) | TTA-IQA · LIVE-IQA DMOS>70 | SROCC | Improved |
| rank loss versus GC loss on batches of similar-quality images (ablation) | TTA-IQA · LIVE-IQA DMOS 28-44 | SROCC | Improved |
| choose single distortion type with maximal predicted quality gap instead of summing all three (ablation) | rank loss · LIVE-IQA | SROCC: 0.634 -> 0.671 | Improved |
| vary GC group fraction p from 0.25 to 0.5 (ablation) | GC loss · KonIQ-10k | SROCC: 0.6516 -> 0.6467 | No clear effect |
| increase number of groups G from 2 to 4 (ablation) | GC loss · KonIQ-10k | SROCC: 0.6516 -> 0.6509 | No clear effect |
| run more than 4 TTA iterations (ablation) | TTA-IQA · CID2013 | SROCC | Hurt |
Closest papers by meaning
- 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 0.87)
- Toward Generalized Image Quality Assessment: Relaxing the Perfect Reference Quality Assumption (CVPR 2025 · similarity 0.86)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 2025 · similarity 0.86)
- Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare (NeurIPS 2024 · similarity 0.86)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 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.
