Image Quality Assessment on KADID
77.5SRCCQ-Hawkeye
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Q-HawkeyeMethodology=MLLM-Based Methods2026.01 | 77.5 | 77.9 | |
| Q-InsightCategory=Specialized, Evaluation Domain=Cross-domain, Evaluation Protocol=Single-task2026.03 | 74.2 | 73.6 | |
| TATARCategory=Unified2026.03 | 73.1 | 72.7 | |
| VisualQuality-R1Venue=NeurIPS’25, Methodology=MLLM-Based Methods2026.01 | 71.9 | 72.3 | |
| Q-insightVenue=NeurIPS’25, Methodology=MLLM-Based Methods2026.01 | 70.2 | 70.2 | |
| DeQACategory=Specialized, Evaluation Domain=Cross-domain, Evaluation Protocol=Single-task2026.03 | 69.4 | 68.7 | |
| DeQA-ScoreVenue=CVPR’25, Methodology=MLLM-Based Methods2026.01 | 68.7 | 69.4 | |
| Q-AlignVenue=ICML’24, Methodology=MLLM-Based Methods2026.01 | 68.4 | 67.4 | |
| UniPerceptCategory=Unified, Training Protocol=Retrained under our protocol2026.03 | 68.2 | 68.5 | |
| GPT-4oCategory=Proprietary, Evaluation Domain=Cross-domain2026.03 | 67.7 | 64.6 | |
| Q-AlignCategory=Specialized, Evaluation Domain=Cross-domain, Source=Cited from UniPercept2026.03 | 67.4 | 68.4 | |
| QwenVL-3-32BCategory=Open-Source, Evaluation Domain=Cross-domain2026.03 | 67.3 | 68.2 | |
| Qwen-SFTVenue=arXiv’25, Methodology=MLLM-Based Methods2026.01 | 66.3 | 66.8 | |
| CLIP-IQA+Venue=AAAI’23, Methodology=Non-MLLM Deep-Learning2026.01 | 65.4 | 65.3 | |
| QwenVL-2.5-72BCategory=Open-Source, Evaluation Domain=Cross-domain2026.03 | 60.6 | 57 | |
| Q-InsightCategory=Specialized, Evaluation Domain=Cross-domain, Source=Cited from UniPercept2026.03 | 58 | 54.8 | |
| InternVL3-78BCategory=Open-Source, Evaluation Domain=Cross-domain2026.03 | 57.9 | 55.3 | |
| InternVL3.5-38BCategory=Open-Source, Evaluation Domain=Cross-domain2026.03 | 56.8 | 53.7 | |
| MUSIQVenue=ICCV’21, Methodology=Non-MLLM Deep-Learning2026.01 | 55.6 | 57.5 | |
| NIMAVenue=TIP’18, Methodology=Non-MLLM Deep-Learning2026.01 | 53.5 | 53.2 | |
| LLaVA-1.5-8BCategory=Open-Source, Evaluation Domain=Cross-domain2026.03 | 50.5 | 53.4 | |
| DBCNNVenue=ICSIPA’19, Methodology=Non-MLLM Deep-Learning2026.01 | 48.4 | 49.7 | |
| HyperIQAVenue=CVPR’20, Methodology=Non-MLLM Deep-Learning2026.01 | 46.8 | 50.6 | |
| ManIQAVenue=CVPR’22, Methodology=Non-MLLM Deep-Learning2026.01 | 46.5 | 49.9 | |
| Compare2ScoreVenue=NIPS’24, Methodology=MLLM-Based Methods2026.01 | 45.3 | 50 | |
| Gemini-2.5-ProCategory=Proprietary, Evaluation Domain=Cross-domain2026.03 | 43.6 | 27.4 | |
| NIQEVenue=SPL’12, Methodology=Handcrafted2026.01 | 40.5 | 46.8 | |
| BRISQUEVenue=TIP’12, Methodology=Handcrafted2026.01 | 35.6 | 42.9 | |
| Claude-4.5Category=Proprietary, Evaluation Domain=Cross-domain2026.03 | 22.3 | 27.3 | |
| DeQA-ScoreTraining Dataset=KonIQ, SPAQ, KADID, PIPAL2026.01 | 0.961 | 0.963 | |
| DeQADomain Setting=In-domain2025.10 | 0.961 | 0.963 | |
| Q-AlignDomain Setting=In-domain2025.10 | 0.954 | 0.95 | |
| DeQA-ScoreTraining Dataset=KonIQ, SPAQ, KADID2026.01 | 0.953 | 0.955 | |
| CONTRIQUEType=SSL + LR2023.10 | 0.934 | 0.937 | |
| VisualQuality-R1Domain Setting=In-domain2025.10 | 0.92 | 0.918 | |
| RACTDomain Setting=In-domain2025.10 | 0.916 | 0.919 | |
| ARNIQAType=SSL + LR2023.10 | 0.908 | 0.912 | |
| Our-EffNetBackbone=EfficientNet [31], Mode=Training-free2024.12 | 0.907 | 0.905 | |
| DeepDC(TIP25)Feature Domain=Feature comparison2026.04 | 0.905 | 0.896 | |
| Q-ProbeCategory=MLLMs-based2026.01 | 0.901 | 0.892 | |
| DBIQA(TPAMI25)Feature Domain=Feature comparison2026.04 | 0.901 | 0.9 | |
| Our-VGGBackbone=VGG [29], Mode=Training-free2024.12 | 0.899 | 0.898 | |
| DeepCausal(CVPR25)Feature Domain=Feature comparison2026.04 | 0.899 | 0.898 | |
| Causal Disentanglement (Ours)Feature Domain=Causal Disentanglement2026.04 | 0.897 | 0.902 | |
| TOPIQ-FR2024.12 | 0.895 | 0.896 | |
| TOPIQ-FR(TIP24)Feature Domain=Feature comparison2026.04 | 0.895 | 0.896 | |
| DeepJSD(TIP25)Feature Domain=Feature comparison2026.04 | 0.894 | 0.893 | |
| Causal Disentanglement (Ours w/o f)Feature Domain=Causal Disentanglement2026.04 | 0.891 | 0.899 | |
| Our-ResNetBackbone=ResNet-50 [8], Mode=Training-free2024.12 | 0.89 | 0.888 | |
| ADISTSFeature Domain=Feature comparison2026.04 | 0.889 | 0.889 | |
| DeepWSD2024.12 | 0.888 | 0.887 | |
| DeepWSD(MM22)Feature Domain=Feature comparison2026.04 | 0.888 | 0.887 | |
| DISTSTraining Dataset=KADID [19]2024.12 | 0.886 | 0.886 | |
| DISTS(TPAMI20)Feature Domain=Feature comparison, trained on dataset=KADID2026.04 | 0.886 | 0.886 | |
| VSI2024.12 | 0.878 | 0.877 | |
| VSIFeature Domain=Traditional2026.04 | 0.878 | 0.877 | |
| Re-IQAType=SSL + LR2023.10 | 0.872 | 0.885 | |
| UniPerceptModel Category=Ours2025.12 | 0.872 | 0.87 | |
| VisualQuality-R1Category=MLLMs-based2026.01 | 0.871 | 0.821 | |
| VisualQuality-R1Training Dataset=KADID, SPAQ2026.01 | 0.868 | 0.867 | |
| Su et al.Type=Supervised learning2023.10 | 0.866 | 0.874 | |
| PieAPP2024.12 | 0.865 | 0.857 | |
| PieAPP(CVPR18)Feature Domain=Feature comparison2026.04 | 0.865 | 0.857 | |
| TRESType=Supervised learning2023.10 | 0.859 | 0.858 | |
| Q-InsightCategory=MLLMs-based2026.01 | 0.856 | 0.881 | |
| FSIM2024.12 | 0.854 | 0.852 | |
| HyperIQAType=Supervised learning2023.10 | 0.852 | 0.845 | |
| DB-CNNType=Supervised learning2023.10 | 0.851 | 0.856 | |
| GMSD2024.12 | 0.847 | 0.847 | |
| GMSDFeature Domain=Traditional2026.04 | 0.847 | 0.847 | |
| DSDFeature Domain=Feature comparison2026.04 | 0.846 | 0.847 | |
| UnifiedReward-TCategory=MLLMs-based2026.01 | 0.841 | 0.877 | |
| LPIPS2024.12 | 0.837 | 0.838 | |
| Q-AlignCategory=MLLMs-based2026.01 | 0.832 | 0.862 | |
| DeQA-ScoreCategory=MLLMs-based2026.01 | 0.831 | 0.873 | |
| PDLFeature Domain=Feature comparison2026.04 | 0.829 | 0.825 | |
| MS-SSIM2024.12 | 0.826 | 0.824 | |
| MS-SSIMFeature Domain=Traditional2026.04 | 0.826 | 0.824 | |
| NLPD2024.12 | 0.812 | 0.809 | |
| NLPDFeature Domain=Traditional2026.04 | 0.812 | 0.809 | |
| LIQECategory=MLLMs-based2026.01 | 0.809 | 0.817 | |
| Qwen2.5-VL-7BCategory=MLLMs-based2026.01 | 0.787 | 0.806 | |
| Q-HawkeyeTraining Dataset=KonIQ2026.01 | 0.775 | 0.779 | |
| Q-DeepSight†Category=MLLM (w/ reasoning), Training Dataset=KonIQ + DQ-7K2026.04 | 0.772 | 0.754 | |
| Q-InsightDomain Setting=In-domain2025.10 | 0.765 | 0.757 | |
| MANIQACategory=Deep Learning2026.01 | 0.76 | 0.78 | |
| Q-DeepSightCategory=MLLM (w/ reasoning), Training Dataset=KonIQ2026.04 | 0.748 | 0.722 | |
| Q-InsightModel Category=Specialized, Domain=Cross domain2025.12 | 0.742 | 0.736 | |
| Q-insightTraining Dataset=KonIQ, KADIS2026.01 | 0.736 | 0.742 | |
| Q-InsightCategory=MLLM-based, Training Dataset=KonIQ2025.10 | 0.736 | 0.742 | |
| Q-Insight†Category=MLLM (w/ reasoning), Training Dataset=KonIQ + DQ-7K2026.04 | 0.736 | 0.742 | |
| LIQECategory=Non-MLLM Deep-learning, Training Dataset=KonIQ2025.10 | 0.735 | 0.72 | |
| MAD2024.12 | 0.726 | 0.717 | |
| RALICategory=Non-MLLM Deep-learning, Training Dataset=KonIQ2025.10 | 0.725 | 0.723 | |
| QwenVL-3-Instruct-8BModel Category=Open-Source, Domain=Cross domain testing2025.12 | 0.723 | 0.696 | |
| Tool-IQATraining Dataset=KonIQ2026.06 | 0.721 | 0.737 | |
| LPIPS(CVPR18)Feature Domain=Feature comparison2026.04 | 0.72 | 0.7 | |
| DeQA-ScoreTraining Dataset=KonIQ, SPAQ2026.01 | 0.719 | 0.724 | |
| VisualQuality-R1Category=VLM (w/o & w/ reasoning), Trained on KonIQ=true2026.01 | 0.712 | 0.703 | |
| VisualQuality-R1Category=MLLM (w/ reasoning), Training Dataset=KonIQ2026.04 | 0.712 | 0.703 |