Instruction Alignment Evaluation on EBench 18K
92.09SRCCReasonEdit
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ReasonEditMethod Category=MLLM-based / Reward Model, Model Scale=4B2026.05 | 92.09 | 75.52 | 92.91 | |
| LMM4Edit*Method Category=MLLM-based2026.05 | 88.3 | 70.8 | 88.98 | |
| DeepSeekVL2 (small)*Method Category=MLLM-based, Model Scale=small2026.05 | 82.8 | 68.93 | 83.9 | |
| InternVL2.5 (8B)*Method Category=MLLM-based, Model Scale=8B2026.05 | 82.07 | 68.09 | 83.87 | |
| AHIQ*Method Category=Deep learning-based FR IQA2026.05 | 52.49 | 36.79 | 54.52 | |
| Q-Align*Method Category=Deep learning-based NR IQA2026.05 | 49.61 | 36.84 | 49.94 | |
| CVRKD*Method Category=Deep learning-based FR IQA2026.05 | 46.61 | 31.7 | 48.06 | |
| TOPIQ*Method Category=Deep learning-based NR IQA2026.05 | 36.41 | 25.36 | 38.48 | |
| PickScoreMethod Category=Vision-language metrics2026.05 | 36.27 | 24.82 | 32.97 | |
| MANIQA*Method Category=Deep learning-based NR IQA2026.05 | 34.32 | 26.61 | 37.65 | |
| ImageRewardMethod Category=Vision-language metrics2026.05 | 28.75 | 19.78 | 31.98 | |
| VQAScoreMethod Category=Vision-language metrics2026.05 | 28.39 | 18.98 | 26.95 | |
| LPIPS (alex)Method Category=Deep learning-based FR IQA, Backbone=AlexNet2026.05 | 22.22 | 14.89 | 29.92 | |
| GMSDMethod Category=Traditional FR IQA2026.05 | 22.08 | 14.85 | 26.93 | |
| LPIPS (vgg)Method Category=Deep learning-based FR IQA, Backbone=VGG2026.05 | 21.66 | 14.52 | 26.32 | |
| CLIPScoreMethod Category=Vision-language metrics2026.05 | 21.52 | 14.49 | 22.09 | |
| SCSSIMMethod Category=Traditional FR IQA2026.05 | 20.3 | 13.5 | 26.7 | |
| Qwen2-VL (7B)Method Category=MLLM-based, Model Scale=7B2026.05 | 19.37 | 14.27 | 17.2 | |
| SSIMMethod Category=Traditional FR IQA2026.05 | 17.05 | 11.32 | 23.01 | |
| BRISQUEMethod Category=Traditional NR IQA2026.05 | 13.66 | 9.23 | 15.62 | |
| NIQEMethod Category=Traditional NR IQA2026.05 | 11.21 | 7.63 | 10.8 | |
| DIIVINEMethod Category=Traditional NR IQA2026.05 | 4.71 | 3.05 | 11.96 |