In this Jupyter notebook i have created three different PyTorch models for computer vision that have been tested on the opensource FashionMNIST dataset. The last model (model 2) was the most successfull one, reaching a 90% accuracy on the test predictions. This was a convolutional network based on the Tiny VVG architecture. The parameters should be saved to the state_dict, however, to achieve this i first set the learning rate of the optimizer (Stochastic gradient descent) to 0.1 for one round of training followed by another round of training with the learining rate 0.01. I used a basic cross entropy loss for the loss function.