Deepfake Detection Model Compression on Swin Base
31.8Storage Cost (MB)DeFakeQ
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DeFakeQBackbone=Swin-Base, Technology Category=Quantization2026.04 | 31.8 | 90.5 | |
| One-bitBackbone=Swin-Base, Technology Category=Low-Rank Factorization2026.04 | 72.7 | 78.3 | |
| HOLaBackbone=Swin-Base, Technology Category=Low-Rank Factorization2026.04 | 86.6 | 74.1 | |
| X-PrunerBackbone=Swin-Base, Technology Category=Model Pruning2026.04 | 133.9 | 60 | |
| GAC-FASBackbone=Swin-Base, Technology Category=Neural Architecture Search2026.04 | 150.7 | 55 | |
| ADDBackbone=Swin-Base, Technology Category=Knowledge Distillation2026.04 | 230.2 | 31.2 | |
| Ground Truth (GT)Backbone=Swin-Base2026.04 | 334.8 | — |