Blind Image Quality Assessment on CSIQ
96.7SRCCGLIANet
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GLIANetNumber of trainable parameters=8M2026.05 | 96.7 | 97.5 | |
| SHDIQANumber of trainable parameters=24M2026.05 | 96.2 | 97.3 | |
| DBCNN2026.05 | 94.6 | 95.9 | |
| Re-IQANumber of trainable parameters=48M2026.05 | 94.5 | 96 | |
| LIQENumber of trainable parameters=151M2026.05 | 94.3 | 94.6 | |
| HyperIQA2026.05 | 92.3 | 94.2 | |
| TReS2026.05 | 92.2 | 94.2 | |
| QMambaNumber of trainable parameters=50M2026.05 | 91.6 | 93.5 | |
| MetaIQA2026.05 | 89.9 | 90.8 | |
| P2P-BM2026.05 | 89.9 | 90.2 | |
| MUSIQ2026.05 | 87.1 | 89.3 | |
| ILNIQE2026.05 | 82.2 | 86.5 | |
| DeQAMethod Category=MLLM-based: SFT training2026.06 | 0.744 | 0.787 | |
| Q-AlignMethod Category=MLLM-based: SFT training2026.06 | 0.737 | 0.671 | |
| MR-IQAMethod Category=MLLM-based: RL training, Backbone=Qwen3-VL-2B, Reproduction Protocol=Controlled reproduction2026.06 | 0.732 | 0.767 | |
| VQ-R1 (ranking)Method Category=MLLM-based: RL training, Backbone=Qwen3-VL-2B, Reproduction Protocol=Controlled reproduction2026.06 | 0.721 | 0.709 | |
| CLIP-IQA+Method Category=Deep-learning-based2026.06 | 0.719 | 0.772 | |
| MUSIQMethod Category=Deep-learning-based2026.06 | 0.71 | 0.771 | |
| C2ScoreMethod Category=MLLM-based: SFT training2026.06 | 0.705 | 0.735 | |
| NIMAMethod Category=Deep-learning-based2026.06 | 0.649 | 0.695 | |
| Q-Insight (regression)Method Category=MLLM-based: RL training, Backbone=Qwen2.5-VL-7B, Reproduction Protocol=Reported2026.06 | 0.64 | 0.685 | |
| NIQEMethod Category=Hand-crafted2026.06 | 0.628 | 0.718 | |
| MANIQAMethod Category=Deep-learning-based2026.06 | 0.627 | 0.623 | |
| DBCNNMethod Category=Deep-learning-based2026.06 | 0.572 | 0.586 | |
| BRISQUEMethod Category=Hand-crafted2026.06 | 0.556 | 0.74 |