This project compares the performance of three deep learning models on fine-grained bird species classification using the CUB-200-2011 dataset. We evaluated two convolutional neural networks (CNNs) on both full and background-masked images, along with a pre-trained ResNet50 model.

Our findings indicate that traditional CNNs struggle with fine-grained classification due to limited feature extraction capabilities. In contrast, ResNet50, benefiting from pre-training on ImageNet and a more advanced architecture, performed significantly better.

For more details, please check out the project report