This project is an extension of a similar project from the Spring 2024 cohort of Data Science Bootcamp. Our goal was to train a convolutional neural network (CNN) that would ingest images of two bodybuilders and output which was more likely to win in competition. Our data set consists of photos from the National Physique Committee website. Each contestant has a variable number of photos from the competition. We manually selected two photos, one front and one back, from each. We divided the data into training, validation, and test sets. We used ResNet50
model. Our data set consisted of 11,776 pairs. After our model was trained, we measured the loss function and accuracy for the
validation set. Despite varying multiple hyperparameters we were unsuccessful in validating either of the models. We hope to collect bigger dataset in the future to improve performance.