Improving data quality in Medical imaging by predicting the orientation of the image in Microsoft Professional Program for Artificial Intelligence capstone project using Microsoft Custom Vision through API calls. Custom Vision was not good enough, so the final submission with better level of predictions was created with Keras but unfortunately that Jupyter notebook is missing. Sorry. See my BeeClassification repo for more complete Custom Vision notebooks and Keras code which was modified for this project.
I used these sources as starting point for my Python code to create MNIST model first which was transferred to BeeClassification Keras code. Yes, the transfer learning with MNIST model did work with those images perfectly. :) https://elitedatascience.com/keras-tutorial-deep-learning-in-python https://github.com/keras-team/keras/blob/master/examples/mnist_cnn.py