AI-generated image detection on Synthbuster
90.1ACC (Clean)B-Free
Evaluation results
| Method | Links | ||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| B-FreePrm (Trainable Parameters)=86.5M, Backbone Protocol (Frozen vs Fine-tuned)=Fine-tuned, Inference Mode (Single-crop vs 5-crop)=5-crop2026.04 | 90.1 | 98.7 | 84.8 | 97.4 | 82.8 | 96.3 | 80.1 | 94.8 | 85.4 | 98.5 | 84.7 | 97.7 | 83 | 95.5 | 82.4 | 98 | 77 | 92.9 | 82.5 | 96.4 | |
| BaselinePrm (Trainable Parameters)=197K, Backbone Protocol (Frozen vs Fine-tuned)=Frozen, Inference Mode (Single-crop vs 5-crop)=Single-crop2026.04 | 86.2 | 89.1 | 56.8 | 79.8 | 53.3 | 77.7 | 52.3 | 75.3 | 85.3 | 88 | 77.2 | 84.7 | 52.3 | 70.1 | 77 | 84.2 | 57.7 | 68.1 | 64 | 78.5 | |
| DCPTPrm (Trainable Parameters)=197K, Backbone Protocol (Frozen vs Fine-tuned)=Frozen, Inference Mode (Single-crop vs 5-crop)=Single-crop2026.04 | 85.3 | 87.9 | 72.5 | 80.5 | 71.2 | 79.4 | 69.5 | 77.6 | 84.4 | 87.1 | 78.8 | 84 | 59.9 | 71.2 | 80.8 | 83.8 | 68 | 71.4 | 73.1 | 79.4 |