Table 3 The F1-score for different architectures depending on loss functions (Day 3).
From: Wildfire spreading prediction using multimodal data and deep neural network approach
| Â | U-Net | U-Net++ | MA-Net | DeepLabV3 |
|---|---|---|---|---|
BCE loss | \(0.63 \pm 0.007\) | \(0.63 \pm 0.004\) | \(0.63 \pm 0.006\) | \(0.64 \pm 0.005\) |
Dice loss | \(0.65 \pm 0.005\) | \(0.66 \pm 0.007\) | \(0.65 \pm 0.007\) | \(0.66 \pm 0.006\) |
Focal loss | \(0.62 \pm 0.006\) | \(0.64 \pm 0.004\) | \(0.63 \pm 0.004\) | \(0.63 \pm 0.006\) |
Dice + focal loss | \(0.66 \pm 0.004\) | \(0.65 \pm 0.004\) | \(0.67 \pm 0.003\) | \(0.65 \pm 0.005\) |
General dice loss | \(0.65 \pm 0.007\) | \(0.66 \pm 0.005\) | \(0.66 \pm 0.005\) | \(0.66 \pm 0.004\) |
Tversky | \(0.65 \pm 0.005\) | \(0.66 \pm 0.006\) | \(0.66 \pm 0.006\) | \(0.66 \pm 0.005\) |
