Research Article
Accurate Multilevel Classification for Wildlife Images
Table 3
Results for multilevel classification that includes different convolutional backbones, input image resolution, and data augmentation techniques.
| ID | Conv. backbone | Resolution | Data augmentation | Top-1 | Top-5 |
| MA | DenseNet201 | 250 × 250 | Standard | 0.51 | 0.78 | MB | DenseNet201 | 250 × 250 | Central Crop | 0.52 | 0.79 | MC | DenseNet201 | 250 × 250 | Multiscale Crop | 0.51 | 0.78 | MD | DenseNet201 | 300 × 300 | Standard | 0.58 | 0.83 | ME | DenseNet201 | 300 × 300 | Central Crop | 0.59 | 0.84 | MF | DenseNet201 | 300 × 300 | Multiscale Crop | 0.61 | 0.85 | MG | EfficientNetB5 | 250 × 250 | Standard | 0.53 | 0.82 | MH | EfficientNetB5 | 250 × 250 | Central Crop | 0.53 | 0.81 | MI | EfficientNetB5 | 250 × 250 | Multiscale Crop | 0.54 | 0.84 | MJ | EfficientNetB5 | 300 × 300 | Standard | 0.61 | 0.87 | MK | EfficientNetB5 | 300 × 300 | Central Crop | 0.62 | 0.87 | ML | EfficientNetB5 | 300 × 300 | Multiscale Crop | 0.62 | 0.88 |
|
|