Comparative Analysis of Deep Learning Architectures for Fine- Grained Bird Classification

This study compares the performance and computational cost of various state-of-the-art deep learning models for fine-grained bird classification. We evaluate ResNet-50, VGG-16, Inception-v3, EfficientNet-B3, MobileNetv3, ViT, DeiT, and ConvNeXt on the CUB- 200-2011 dataset. Each model was trained using the PyTorch library on a high-performance server, with early stopping employed to prevent overfitting. Our results indicate that DeiT achieves the highest accuracy (85.6%) and F1 score (85.6%), while MobileNetv3 shows the best computational efficiency. These findings offer valuable insights into selecting appropriate models for fine-grained classification tasks based on specific application requirements.