A deep dive into computer vision (CV) for Human Action Recognition (HAR). My goal is to use transfer learning with multiple state of the art computer vision models, tested on ImageNet, and apply them to my specific HAR application. This is to test whether the speed of training from transfer learning outweights the possible biases that can be found within the predefined weights of these models. To keep the testing simple, the trainable layers being added on are only dense and dropout layers. For these trainable layers, I am also using Keras Tuner to explore the full potential that these fully adapted models can obtain. Enjoy the code!