Image classification using PyTorch and Core ML

The aim of the study was to compare different machine learning models trained using the PyTorch library in Python and the Core ML library in the Create ML tool. In the case of PyTorch, using transfer learning on a pre-trained ResNet50 model, data augmentation and normalization, four models were trained on two various data sets, achieving accuracy, precision, recall and F1 score above 80%. Four identical models were trained on the same data sets in the Create ML tool, and the conversion of the PyTorch models to the Core ML format allowed for a reliable comparison. This also emphasizes the effectiveness of conversion using the coremltools library, while maintaining model performance. The study emphasizes the key role of dataset quality and techniques for improving dataset quality.