Explanation Quality Evaluation on Deepfake Detection Dataset DDIM, PixArt, SD, SiT, StyleGAN
4.42CACPRPO
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PRPO2025.09 | 4.42 | 4.56 | 4.58 | 4.5 | 4.69 | 4.55 | |
| Gemini-2.52025.09 | 3.98 | 4.16 | 4.23 | 4.05 | 4.6 | 4.2 | |
| DX-LLaVA2025.09 | 3.78 | 3.99 | 4.04 | 3.98 | 4.29 | 4.02 | |
| Gemma-3-27B-IT2025.09 | 3.36 | 3.77 | 3.76 | 3.51 | 4.42 | 3.76 | |
| Pixtral-12B2025.09 | 3.17 | 3.51 | 3.4 | 3.38 | 4.2 | 3.53 | |
| LLaVA-Base2025.09 | 3.07 | 2.76 | 2.92 | 3.18 | 3.94 | 3.17 | |
| Qwen2.5-VL-32B2025.09 | 3.02 | 3.7 | 3.64 | 3.28 | 4.29 | 3.59 | |
| GPT-4o-Mini2025.09 | 2.98 | 1.32 | 1.94 | 2.67 | 2.99 | 2.38 | |
| LLaMA-4-Maverick2025.09 | 2.7 | 3.34 | 3.26 | 3.01 | 4.06 | 3.27 |