No-Reference Image Quality Assessment on SPAQ (test)
0.924SROCCQ-Tacit†
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Q-Tacit†Method Category=VLM-based, Training Protocol=Multi-task training set in [25] for GRPO calibration2026.03 | 0.924 | 0.913 | |
| Q-TacitMethod Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.923 | 0.911 | |
| HyperIQAModel Type=Supervised pretraining and supervised fine-tuning2021.10 | 0.916 | 0.919 | |
| CONTRIQUEModel Type=Unsupervised pretraining and Linear Regression2021.10 | 0.914 | 0.919 | |
| DB-CNNModel Type=Supervised pretraining and supervised fine-tuning2021.10 | 0.911 | 0.915 | |
| CONTRIQUEOpinion-Unaware=false2024.03 | 0.91 | 0.912 | |
| Re-IQAOpinion-Unaware=false2024.03 | 0.909 | 0.912 | |
| Q-Insight†Method Category=VLM-based, Training Protocol=Multi-task training set in [25] for GRPO calibration2026.03 | 0.905 | 0.907 | |
| ResNet-50Model Type=Supervised pretraining and Linear Regression2021.10 | 0.904 | 0.909 | |
| ARNIQAOpinion-Unaware=false2024.03 | 0.904 | 0.909 | |
| GRepQOpinion-Unaware=false2024.03 | 0.902 | 0.903 | |
| CLIP-IQA+Opinion-Unaware=false2024.03 | 0.901 | 0.904 | |
| Q-InsightMethod Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.899 | 0.903 | |
| DeQA-ScoreMethod Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.896 | 0.895 | |
| VisualQuality-R1Method Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.895 | 0.894 | |
| Q-AlignMethod Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.887 | 0.886 | |
| Qwen-2.5-VL-7BMethod Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.875 | 0.874 | |
| CLIP-IQA+Method Category=Non-VLM Deep-learning, Training Protocol=Trained on KonIQ-10K2026.03 | 0.864 | 0.866 | |
| MUSIQMethod Category=Non-VLM Deep-learning, Training Protocol=Trained on KonIQ-10K2026.03 | 0.863 | 0.868 | |
| Compare2ScoreMethod Category=VLM-based, Training Protocol=Trained on KonIQ-10K2026.03 | 0.86 | 0.867 | |
| NIMAMethod Category=Non-VLM Deep-learning, Training Protocol=Trained on KonIQ-10K2026.03 | 0.856 | 0.838 | |
| HOSAModel Type=Codebook-based Features2021.10 | 0.846 | 0.852 | |
| QualiCLIPOpinion-Unaware=true2024.03 | 0.841 | 0.851 | |
| BRISQUEModel Type=Traditional/Handcrafted Features2021.10 | 0.809 | 0.817 | |
| MDFSOpinion-Unaware=true2024.03 | 0.805 | 0.812 | |
| GRepQ-OUOpinion-Unaware=true2024.03 | 0.802 | 0.805 | |
| ARNIQA-OUOpinion-Unaware=true2024.03 | 0.788 | 0.797 | |
| ManIQAMethod Category=Non-VLM Deep-learning, Training Protocol=Trained on KonIQ-10K2026.03 | 0.758 | 0.768 | |
| CLIP-IQAOpinion-Unaware=true2024.03 | 0.733 | 0.734 | |
| IL-NIQEOpinion-Unaware=true2024.03 | 0.713 | 0.724 | |
| CORNIAModel Type=Codebook-based Features2021.10 | 0.709 | 0.725 | |
| CL-MIOpinion-Unaware=true2024.03 | 0.701 | 0.701 | |
| NIQEModel Type=Traditional/Handcrafted Features2021.10 | 0.7 | 0.709 | |
| NIQEOpinion-Unaware=true2024.03 | 0.7 | 0.712 | |
| CONTRIQUE-OUOpinion-Unaware=true2024.03 | 0.677 | 0.685 | |
| NIQEMethod Category=Handcrafted, Training Protocol=Trained on KonIQ-10K2026.03 | 0.664 | 0.679 | |
| Re-IQA-OUOpinion-Unaware=true2024.03 | 0.613 | 0.616 | |
| BRISQUEMethod Category=Handcrafted, Training Protocol=Trained on KonIQ-10K2026.03 | 0.406 | 0.49 |