AI Image Detection on Stable Diffusion (test)
100AccuracyMobilenetV3 + DIRE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MobilenetV3 + DIRETrain All Layers=True, Trainable Parameters=4M2026.06 | 100 | — | — | — | |
| MobilenetV3Train All Layers=True, Trainable Parameters=4M2026.06 | 99.9 | — | — | — | |
| PSO-based Rejection MechanismBackbone=ResNet50, Evaluation Protocol=In-distribution (Same technique)2025.12 | 99.79 | 99.76 | 99.82 | 99.79 | |
| EfficientNetV2Train All Layers=True, Trainable Parameters=22.1M2026.06 | 99.6 | — | — | — | |
| Resnet50Train All Layers=True, Trainable Parameters=23.5M2026.06 | 99.2 | — | — | — | |
| Swin-tinyTrain All Layers=True, Trainable Parameters=27.5M2026.06 | 99.1 | — | — | — | |
| MobilenetV3 + DIRETrain All Layers=False, Trainable Parameters=25622026.06 | 97.2 | — | — | — | |
| ViT-smallTrain All Layers=True, Trainable Parameters=22.4M2026.06 | 96.3 | — | — | — | |
| Resnet50Train All Layers=False, Trainable Parameters=40982026.06 | 94 | — | — | — | |
| MobilenetV3Train All Layers=False, Trainable Parameters=25622026.06 | 91.2 | — | — | — | |
| Swin-tinyTrain All Layers=False, Trainable Parameters=15382026.06 | 91.2 | — | — | — | |
| ViT-smallTrain All Layers=False, Trainable Parameters=7702026.06 | 90.5 | — | — | — | |
| EfficientNetV2Train All Layers=False, Trainable Parameters=35862026.06 | 89.5 | — | — | — |