Deepfake Detection on COCOFake (val)
99.68AccuracyOpenCLIP-ViT-B/32
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OpenCLIP-ViT-B/32Model=OpenCLIP-ViT-B/32, Pre-training dataset=LAION-2B, Parameters (M)=88.3, FLOPs (G)=8.562024.06 | 99.68 | — | |
| BNext-SModel=BNext-S, Backbone=trainable, Pre-training dataset=ImageNet, Parameters (M)=67.1, FLOPs (G)=1.912024.06 | 99.28 | 99.89 | |
| BNext-TModel=BNext-T, Backbone=trainable, Pre-training dataset=ImageNet, Parameters (M)=29.8, FLOPs (G)=0.892024.06 | 99.25 | 99.86 | |
| BNext-MModel=BNext-M, Backbone=trainable, Pre-training dataset=ImageNet, Parameters (M)=133, FLOPs (G)=3.392024.06 | 99.18 | 99.91 | |
| CLIP-VIT-B/32Model=CLIP-VIT-B/32, Pre-training dataset=OpenAI WIT, Parameters (M)=88.3, FLOPs (G)=8.562024.06 | 99.11 | — | |
| CLIP-ResNet50Model=CLIP-ResNet50, Pre-training dataset=OpenAI WIT, Parameters (M)=25.6, FLOPs (G)=4.82024.06 | 99.07 | — | |
| OpenCLIP-ViT-B/32Model=OpenCLIP-ViT-B/32, Pre-training dataset=LAION-400M, Parameters (M)=88.3, FLOPs (G)=8.562024.06 | 97.88 | — | |
| BNext-SModel=BNext-S, Backbone=frozen, Pre-training dataset=ImageNet, Parameters (M)=67.1, FLOPs (G)=1.912024.06 | 93.15 | 95.19 | |
| ResNet50Model=ResNet50, Pre-training dataset=ImageNet, Parameters (M)=25.6, FLOPs (G)=4.82024.06 | 90.31 | — | |
| ViT-B/32Model=ViT-B/32, Pre-training dataset=ImageNet, Parameters (M)=88.3, FLOPs (G)=8.562024.06 | 87.64 | — | |
| BNext-MModel=BNext-M, Backbone=frozen, Pre-training dataset=ImageNet, Parameters (M)=133, FLOPs (G)=3.392024.06 | 84.59 | 82.11 | |
| BNext-TModel=BNext-T, Backbone=frozen, Pre-training dataset=ImageNet, Parameters (M)=29.8, FLOPs (G)=0.892024.06 | 83.65 | 81.98 |