Yeah, his baldness detector is not robust as it just checks the average color of the area above the forehead, but you could make a robust one through transfer l

Yeah, his baldness detector is not robust as it just checks the average color of the area above the forehead, but you could make a robust one through transfer learning a deep model trained on ImageNet further finetuned on a small dataset of a few dozen images of bald and non-bald. Something like this: https://www.kaggle.com/dansbecker/transfer-learning https://www.kaggle.com/dansbecker/transfer-learning This method gets good, robust performance with just a few minutes of training.