VGG, MobileNet and AlexNet on Recognizing Skin Cancer Symptoms

Skin cancer is one of the most common cancers. Using convolutional neural network to compare diagonalize skin cancer has become popular in recent years. Yet, comparison across different models has been difficult due to the variety of datasets used by the models' creators and problems such as over-fitting. In this work, I compared the performance across these most classic CNN models including VGG-16, VGG-19, MobileNet, ResNet by training and testing them on the same dataset and compare their performance. In fact, this comparison is meaningful and necessary since it can tell us which model can do well in this dataset and it will help us to further study the nature of structure in different models. My comparison concludes that when trained with identical batch sizes, the models' accuracy and loss have little variety, and VGG-16 has the highest accuracy and minimum loss.