Perceptual Quality Assessment on HPE-Bench 1.0 (test)
0.887SRCCLayer-Selective MLLM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Layer-Selective MLLMcategory=Our Framework2026.01 | 0.887 | 0.7668 | 0.8959 | |
| AHIQcategory=Deep Learning-based FR IQA, reference_type=FR, fine-tuned=true2026.01 | 0.8824 | 0.6578 | 0.8805 | |
| Layer-Selective MLLMcategory=Our Framework2026.01 | 0.8687 | 0.743 | 0.8852 | |
| Q-Aligncategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.8527 | 0.5825 | 0.8793 | |
| DeepSeekVL2 (small)category=LMM-Based Models, fine-tuned=true2026.01 | 0.8399 | 0.7137 | 0.8321 | |
| Qwen3-VL (8B)category=LMM-Based Models, fine-tuned=true2026.01 | 0.8377 | 0.652 | 0.8331 | |
| Qwen3-VL (8B)category=LMM-Based Models, fine-tuned=true2026.01 | 0.8336 | 0.7072 | 0.8587 | |
| LLaVA-NeXT (8B)category=LMM-Based Models, fine-tuned=true2026.01 | 0.8161 | 0.6634 | 0.8249 | |
| DeepSeekVL2 (small)category=LMM-Based Models, fine-tuned=true2026.01 | 0.8054 | 0.643 | 0.8273 | |
| TOPIQcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.8026 | 0.618 | 0.7919 | |
| MANIQAcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.7955 | 0.6681 | 0.8454 | |
| CVRKDcategory=Deep Learning-based FR IQA, reference_type=FR, fine-tuned=true2026.01 | 0.7827 | 0.6094 | 0.8314 | |
| LLaVA-NeXT (8B)category=LMM-Based Models, fine-tuned=true2026.01 | 0.7821 | 0.6603 | 0.8071 | |
| CVRKDcategory=Deep Learning-based FR IQA, reference_type=FR, fine-tuned=true2026.01 | 0.774 | 0.6128 | 0.8807 | |
| AHIQcategory=Deep Learning-based FR IQA, reference_type=FR, fine-tuned=true2026.01 | 0.7528 | 0.6367 | 0.7575 | |
| Qwen3-VL (7B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.7436 | 0.5638 | 0.7456 | |
| DBCNNcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.7294 | 0.608 | 0.7997 | |
| Q-Aligncategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.7119 | 0.4738 | 0.7852 | |
| HyperIQAcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.711 | 0.4972 | 0.6919 | |
| LPIPScategory=Deep Learning-based FR IQA, reference_type=FR2026.01 | 0.6867 | 0.5678 | 0.7111 | |
| MANIQAcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.6372 | 0.4653 | 0.6415 | |
| DBCNNcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.6074 | 0.4255 | 0.6482 | |
| Qwen3-VL (7B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.5822 | 0.4427 | 0.509 | |
| FSIMcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.5664 | 0.4067 | 0.5636 | |
| TOPIQcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.5185 | 0.379 | 0.5638 | |
| MSEcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.4996 | 0.3414 | 0.5415 | |
| SSIMcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.4961 | 0.3495 | 0.4519 | |
| InternVL3 (8B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.4629 | 0.3655 | 0.2437 | |
| ST-LPIPScategory=Deep Learning-based FR IQA, reference_type=FR2026.01 | 0.4575 | 0.3128 | 0.4273 | |
| PSNRcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.4342 | 0.3566 | 0.4508 | |
| ImageRewardcategory=Vision-Language Methods2026.01 | 0.4304 | 0.295 | 0.4392 | |
| ImageRewardcategory=Vision-Language Methods2026.01 | 0.4288 | 0.2588 | 0.3997 | |
| BRISQUEcategory=Traditional NR IQA, reference_type=NR2026.01 | 0.3699 | 0.2432 | 0.3868 | |
| MiniCPM-V2.6 (8B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.3581 | 0.2851 | 0.3403 | |
| BIQIcategory=Traditional NR IQA, reference_type=NR2026.01 | 0.3182 | 0.1867 | 0.3335 | |
| LLaVAScorecategory=Vision-Language Methods2026.01 | 0.3154 | 0.1934 | 0.3668 | |
| LLaVAScorecategory=Vision-Language Methods2026.01 | 0.3085 | 0.2015 | 0.4258 | |
| InternVL3 (8B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.3053 | 0.2529 | 0.0002 | |
| HyperIQAcategory=Deep Learning-based NR IQA, reference_type=NR, fine-tuned=true2026.01 | 0.2978 | 0.1902 | 0.3063 | |
| VQAScorecategory=Vision-Language Methods2026.01 | 0.2796 | 0.215 | 0.325 | |
| DeepSeekVL2 (small)category=LMM-Based Models, fine-tuned=false2026.01 | 0.2788 | 0.2266 | 0.119 | |
| MiniCPM-V2.6 (8B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.2706 | 0.2198 | 0.3094 | |
| PickScorecategory=Vision-Language Methods2026.01 | 0.2586 | 0.1562 | 0.2683 | |
| DeepSeekVL2 (small)category=LMM-Based Models, fine-tuned=false2026.01 | 0.2577 | 0.2104 | 0.212 | |
| CLIPScorecategory=Vision-Language Methods2026.01 | 0.2446 | 0.1526 | 0.2597 | |
| VQAScorecategory=Vision-Language Methods2026.01 | 0.2267 | 0.1429 | 0.2407 | |
| CLIPScorecategory=Vision-Language Methods2026.01 | 0.2024 | 0.1326 | 0.2417 | |
| mPLUG-Owl3 (7B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.2002 | 0.1383 | 0.0144 | |
| LPIPScategory=Deep Learning-based FR IQA, reference_type=FR2026.01 | 0.1903 | 0.1351 | 0.3024 | |
| mPLUG-Owl3 (7B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.1802 | 0.1289 | 0.1303 | |
| BIQIcategory=Traditional NR IQA, reference_type=NR2026.01 | 0.1623 | 0.0968 | 0.252 | |
| LLaVA-NeXT (8B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.1591 | 0.1298 | 0.0964 | |
| DIIVINEcategory=Traditional NR IQA, reference_type=NR2026.01 | 0.1541 | 0.0918 | 0.3639 | |
| BRISQUEcategory=Traditional NR IQA, reference_type=NR2026.01 | 0.1424 | 0.0805 | 0.1945 | |
| LLaVA-1.5 (7B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.1402 | 0.1149 | 0.1189 | |
| PickScorecategory=Vision-Language Methods2026.01 | 0.1359 | 0.0929 | 0.1977 | |
| LLama3.2-Vision (11B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.0967 | 0.0785 | 0.1088 | |
| LLaVA-NeXT (8B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.0818 | 0.0675 | 0.0131 | |
| LLama3.2-Vision (11B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.0818 | 0.0675 | 0.0131 | |
| LLaVA-1.5 (7B)category=LMM-Based Models, fine-tuned=false2026.01 | 0.0746 | 0.0611 | 0.0642 | |
| FSIMcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.0508 | 0.0326 | 0.2242 | |
| MSEcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.0268 | 0.0191 | 0.2215 | |
| PSNRcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.0245 | 0.0169 | 0.2228 | |
| DIIVINEcategory=Traditional NR IQA, reference_type=NR2026.01 | 0.0128 | 0.0081 | 0.1904 | |
| ST-LPIPScategory=Deep Learning-based FR IQA, reference_type=FR2026.01 | 0.0054 | 0.0045 | 0.0482 | |
| SSIMcategory=Traditional FR IQA, reference_type=FR2026.01 | 0.0038 | 0.0007 | 0.2207 |