Spatial Aesthetic Image Quality Assessment on SA-BENCH
0.831Layout PLCCGLM-4-1V-9B
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| GLM-4-1V-9BMethod Category=MLLMs (SFT-based)2025.12 | 0.831 | 0.821 | 0.848 | 0.834 | 0.69 | 0.69 | 0.532 | 0.503 | 0.836 | 0.842 | |
| SA-IQAMethod Category=MLLMs (SFT-based)2025.12 | 0.831 | 0.822 | 0.896 | 0.895 | 0.724 | 0.694 | 0.657 | 0.596 | 0.864 | 0.86 | |
| Qwen3-VL-4BMethod Category=MLLMs (SFT-based)2025.12 | 0.821 | 0.807 | 0.892 | 0.891 | 0.712 | 0.701 | 0.558 | 0.488 | 0.853 | 0.849 | |
| Qwen3-VL-8BMethod Category=MLLMs (SFT-based)2025.12 | 0.814 | 0.804 | 0.892 | 0.889 | 0.696 | 0.705 | 0.529 | 0.506 | 0.849 | 0.846 | |
| Ovis2-5-2BMethod Category=MLLMs (SFT-based)2025.12 | 0.812 | 0.807 | 0.89 | 0.888 | 0.692 | 0.681 | 0.556 | 0.52 | 0.848 | 0.844 | |
| Qwen3-VL-2BMethod Category=MLLMs (SFT-based)2025.12 | 0.808 | 0.802 | 0.886 | 0.884 | 0.684 | 0.687 | 0.481 | 0.439 | 0.841 | 0.836 | |
| mPLUG-Owl3-7BMethod Category=MLLMs (SFT-based)2025.12 | 0.784 | 0.773 | 0.831 | 0.822 | 0.662 | 0.675 | 0.456 | 0.421 | 0.81 | 0.81 | |
| InternVL3-5-8BMethod Category=MLLMs (SFT-based)2025.12 | 0.774 | 0.769 | 0.885 | 0.882 | 0.702 | 0.693 | 0.539 | 0.462 | 0.837 | 0.829 | |
| Q-Eval-ScoreMethod Category=Deep Learning-based NR-IQA2025.12 | 0.265 | 0.266 | 0.149 | 0.142 | 0.036 | 0.042 | 0.144 | 0.153 | 0.162 | 0.168 | |
| gemini-2.5-proMethod Category=Commercial MLLMs2025.12 | 0.246 | 0.181 | 0.594 | 0.603 | -0.023 | -0.043 | 0.252 | 0.225 | 0.414 | 0.393 | |
| DeQA-ScoreMethod Category=Deep Learning-based NR-IQA2025.12 | 0.201 | 0.174 | 0.139 | 0.105 | 0.305 | 0.351 | 0.025 | 0.006 | 0.111 | 0.141 | |
| CLIPIQAMethod Category=Traditional NR-IQA2025.12 | 0.197 | 0.23 | 0.097 | 0.101 | 0.062 | 0.084 | 0.113 | 0.128 | 0.118 | 0.135 | |
| MANIQAMethod Category=Traditional NR-IQA2025.12 | 0.194 | 0.21 | 0.096 | 0.105 | 0.125 | 0.147 | 0.071 | 0.073 | 0.118 | 0.139 | |
| gpt-5Method Category=Commercial MLLMs2025.12 | 0.185 | 0.147 | 0.483 | 0.49 | -0.031 | -0.054 | 0.204 | 0.177 | 0.318 | 0.308 | |
| HyperIQAMethod Category=Traditional NR-IQA2025.12 | 0.179 | 0.217 | 0.135 | 0.136 | 0.161 | 0.181 | 0.163 | 0.131 | 0.144 | 0.165 | |
| DBCNNMethod Category=Traditional NR-IQA2025.12 | 0.162 | 0.173 | -0.006 | -0.007 | 0.198 | 0.197 | 0.105 | 0.096 | 0.082 | 0.1 | |
| Compare2ScoreMethod Category=Deep Learning-based NR-IQA2025.12 | 0.16 | 0.162 | 0.074 | 0.075 | 0.274 | 0.288 | 0.028 | 0.011 | 0.088 | 0.115 | |
| Q-SiTMethod Category=Deep Learning-based NR-IQA2025.12 | 0.141 | 0.19 | 0.098 | 0.077 | 0.207 | 0.219 | 0.184 | 0.131 | 0.115 | 0.14 | |
| MUSIQMethod Category=Traditional NR-IQA2025.12 | 0.121 | 0.138 | 0.007 | -0.017 | 0.133 | 0.228 | 0.019 | -0.001 | 0.057 | 0.084 | |
| qwen-vl-maxMethod Category=Commercial MLLMs2025.12 | 0.088 | 0.067 | 0.526 | 0.528 | -0.037 | -0.058 | 0.091 | 0.078 | 0.362 | 0.359 | |
| Q-Align-aestheticsMethod Category=Deep Learning-based NR-IQA2025.12 | 0.075 | 0.082 | 0.318 | 0.32 | 0.116 | 0.12 | 0.001 | 0.026 | 0.133 | 0.142 | |
| Q-Align-qualityMethod Category=Deep Learning-based NR-IQA2025.12 | 0.057 | 0.101 | 0.183 | 0.187 | 0.299 | 0.294 | 0.088 | 0.074 | 0.11 | 0.128 | |
| Q-InsightMethod Category=Deep Learning-based NR-IQA2025.12 | 0.053 | 0.06 | 0.08 | 0.093 | 0.286 | 0.286 | 0.067 | 0.032 | 0.083 | 0.089 | |
| gpt-4oMethod Category=Commercial MLLMs2025.12 | 0.017 | -0.009 | 0.537 | 0.542 | 0.064 | 0.038 | 0.233 | 0.149 | 0.312 | 0.298 | |
| claude3.5-sonnetMethod Category=Commercial MLLMs2025.12 | -0.001 | -0.032 | 0.542 | 0.552 | -0.179 | -0.197 | 0.058 | -0.006 | 0.326 | 0.304 |