Table 5 Simulation hyperparameters of existing methods.
From: A gated temporal attention based intra prediction framework for robust deepfake video detection
S.No | Method | Parameter | Type or Value |
|---|---|---|---|
1 | XceptionNet | Input Size | 299 × 299 × 3 |
2 | Backbone | Depthwise Separable ConvNet | |
3 | Learning Rate | 0.0001 | |
4 | Batch Size | 32 | |
5 | Epochs for Training | 100 | |
6 | Loss Function | Binary Crossentropy | |
7 | Two-Stream CNN | Input Size | 224 × 224 × 3 per stream |
8 | Backbone | VGG-16 for both streams | |
9 | Loss Function | Categorical Crossentropy | |
10 | Batch Size | 32 | |
11 | Epochs for Training | 100 | |
12 | Learning Rate | 0.0001 | |
13 | EfficientNet-B0 | Input Size | 224 × 224 × 3 |
14 | Input Size | 128 × 128 × 3 | |
15 | Backbone | MBConv with squeeze-excite | |
16 | Dropout Rate | 0.2 | |
17 | Loss Function | Sparse Categorical Loss | |
18 | Learning Rate | 0.0001 | |
19 | Batch Size | 32 | |
20 | Epochs for Training | 100 | |
21 | Capsule Network | Capsule Routing Iterations | 3 |
22 | Optimizer for All Models | Adam | |
23 | Initial Learning Rate | 0.001 | |
24 | Batch Size (All Models) | 32 | |
25 | Epochs for Training | 100 | |
26 | Dropout Rate | 0.3 | |
27 | Loss Function | Margin Loss |
