Image Quality Assessment on SPAQ (test)
0.93SRCCQ-Align
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Q-Aligninput_modality=image-only2025.07 | 0.93 | 0.933 | |
| RvTCinput_modality=image-only2025.07 | 0.926 | 0.93 | |
| LIQEinput_modality=image-only2025.07 | 0.922 | 0.919 | |
| MUSIQinput_modality=image-only2025.07 | 0.917 | 0.921 | |
| NIMAinput_modality=image-only2025.07 | 0.907 | 0.91 | |
| OursCategory=SFT-based and RL-based MLLMs, Condition=Image-conditioned2025.12 | 0.907 | 0.893 | |
| NTPCoT training=true2025.03 | 0.904 | 0.906 | |
| Q-Insight-ScoreModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.899 | 0.903 | |
| Q-Insight-ScoreCategory=SFT-based and RL-based MLLMs, Condition=Image-conditioned2025.12 | 0.899 | 0.887 | |
| DeQAModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.896 | 0.895 | |
| DeQACategory=SFT-based and RL-based MLLMs, Condition=Image-conditioned2025.12 | 0.896 | 0.895 | |
| Q-AlignCoT training=false2025.03 | 0.887 | 0.886 | |
| Q-AlignModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.887 | 0.886 | |
| Q-AlignCategory=SFT-based and RL-based MLLMs, Condition=Image-conditioned2025.12 | 0.887 | 0.886 | |
| Chain-of-ThoughtModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.884 | 0.886 | |
| Self-ConsistencyModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.882 | 0.883 | |
| Autoencoder-likeModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.882 | 0.884 | |
| LIQEtraining=jointly trained on six datasets2023.03 | 0.881 | — | |
| NTPCoT training=false2025.03 | 0.878 | 0.875 | |
| OursCategory=Caption-Only Conditions, Condition=Caption-Only2025.12 | 0.875 | 0.861 | |
| DEFNetZero-shot=true2025.07 | 0.868 | — | |
| CLIP-IQA+CoT training=false2025.03 | 0.864 | 0.866 | |
| CLIP-IQA+Model category=Deep-Learning Models, Input Condition=Image-conditioned2026.01 | 0.864 | 0.866 | |
| CLIP-IQA+Category=Deep-Learning Models, Condition=Image-conditioned2025.12 | 0.864 | 0.866 | |
| MUSIQModel category=Deep-Learning Models, Input Condition=Image-conditioned2026.01 | 0.863 | 0.868 | |
| MUSIQCategory=Deep-Learning Models, Condition=Image-conditioned2025.12 | 0.863 | 0.868 | |
| Self-ConsistencyInput Condition=Text-Only2026.01 | 0.861 | 0.864 | |
| C2ScoreModel category=SFT-based and RL-based MLLMs, Input Condition=Image-conditioned2026.01 | 0.86 | 0.867 | |
| C2ScoreCategory=SFT-based and RL-based MLLMs, Condition=Image-conditioned2025.12 | 0.86 | 0.867 | |
| Self-ConsistencyInput Condition=Text-Only, Text Filtering=Score-related words removed2026.01 | 0.859 | 0.861 | |
| NIMAModel category=Deep-Learning Models, Input Condition=Image-conditioned2026.01 | 0.856 | 0.838 | |
| NIMACategory=Deep-Learning Models, Condition=Image-conditioned2025.12 | 0.856 | 0.838 | |
| MUSIQ2023.03 | 0.853 | — | |
| MUSIQZero-shot=true2025.07 | 0.853 | — | |
| Q-Insight-ScoreCategory=Caption-Only Conditions, Condition=Caption-Only2025.12 | 0.847 | 0.84 | |
| Qwen3-VL-8B-InstructRatio=1:9, IQARAG integration=false2026.01 | 0.8427 | 0.792 | |
| Qwen3-VL-8B-InstructRatio=3:7, IQARAG integration=false2026.01 | 0.8427 | 0.7944 | |
| Kimi-VL-A3B-InstructRatio=3:7, IQARAG integration=true2026.01 | 0.8427 | 0.8687 | |
| Qwen3-VL-8B-InstructRatio=1:4, IQARAG integration=false2026.01 | 0.8415 | 0.79 | |
| Autoencoder-likeInput Condition=Text-Only2026.01 | 0.839 | 0.824 | |
| UNIQUEtraining=jointly trained on six datasets2023.03 | 0.838 | — | |
| Autoencoder-likeInput Condition=Text-Only, Text Filtering=Score-related words removed2026.01 | 0.838 | 0.834 | |
| UNIQUEZero-shot=true2025.07 | 0.838 | — | |
| LIQECoT training=false2025.03 | 0.833 | 0.846 | |
| Q-Insight-ScoreInput Condition=Text-Only2026.01 | 0.833 | 0.832 | |
| Chain-of-ThoughtInput Condition=Text-Only2026.01 | 0.833 | 0.829 | |
| Q-Insight-ScoreInput Condition=Text-Only, Text Filtering=Score-related words removed2026.01 | 0.832 | 0.831 | |
| Chain-of-ThoughtInput Condition=Text-Only, Text Filtering=Score-related words removed2026.01 | 0.831 | 0.831 | |
| Kimi-VL-A3B-InstructRatio=1:9, IQARAG integration=false2026.01 | 0.8283 | 0.8622 | |
| Kimi-VL-A3B-InstructRatio=1:4, IQARAG integration=false2026.01 | 0.8281 | 0.8619 | |
| Kimi-VL-A3B-InstructRatio=3:7, IQARAG integration=false2026.01 | 0.8266 | 0.8619 | |
| Kimi-VL-A3B-InstructRatio=1:9, IQARAG integration=true2026.01 | 0.8249 | 0.8552 | |
| PaQ2PiQ2023.03 | 0.823 | — | |
| PaQ2PiQZero-shot=true2025.07 | 0.823 | — | |
| Qwen3-VL-8B-InstructRatio=3:7, IQARAG integration=true2026.01 | 0.8188 | 0.8273 | |
| Qwen3-VL-8B-InstructRatio=1:4, IQARAG integration=true2026.01 | 0.8183 | 0.8307 | |
| Kimi-VL-A3B-InstructRatio=1:4, IQARAG integration=true2026.01 | 0.8177 | 0.8509 | |
| Qwen3-VL-8B-InstructRatio=1:9, IQARAG integration=true2026.01 | 0.8169 | 0.8282 | |
| DBCNNModel category=Deep-Learning Models, Input Condition=Image-conditioned2026.01 | 0.806 | 0.812 | |
| DBCNNCategory=Deep-Learning Models, Condition=Image-conditioned2025.12 | 0.806 | 0.812 | |
| DBCNNTraining Dataset=KonIQ-10K2023.03 | 0.801 | — | |
| DBCNNZero-shot=true, Training Source=KADID-10k2025.07 | 0.801 | — | |
| InternVL3.5-8BRatio=3:7, IQARAG integration=true2026.01 | 0.7932 | 0.8006 | |
| HyperIQACoT training=false2025.03 | 0.788 | 0.791 | |
| InternVL3.5-8BRatio=1:4, IQARAG integration=true2026.01 | 0.7841 | 0.8019 | |
| InternVL3.5-8BRatio=1:9, IQARAG integration=true2026.01 | 0.7631 | 0.7672 | |
| MANIQAModel category=Deep-Learning Models, Input Condition=Image-conditioned2026.01 | 0.758 | 0.768 | |
| MANIQACategory=Deep-Learning Models, Condition=Image-conditioned2025.12 | 0.758 | 0.768 | |
| NIQECategory=Hand-Crafted Models, Condition=Image-conditioned2025.12 | 0.664 | 0.679 | |
| InternVL3.5-8BRatio=1:9, IQARAG integration=false2026.01 | 0.6479 | 0.6711 | |
| InternVL3.5-8BRatio=3:7, IQARAG integration=false2026.01 | 0.6469 | 0.6709 | |
| InternVL3.5-8BRatio=1:4, IQARAG integration=false2026.01 | 0.6452 | 0.6691 | |
| NIQE2023.03 | 0.578 | — | |
| NIQEZero-shot=true2025.07 | 0.578 | — | |
| DBCNNTraining Dataset=KADID-10K2023.03 | 0.412 | — | |
| DBCNNZero-shot=true, Training Source=KonIQ-10k2025.07 | 0.412 | — | |
| BRISQUECategory=Hand-Crafted Models, Condition=Image-conditioned2025.12 | 0.406 | 0.49 |