This repository is the code I developed to test the impact of using imperfectly annotated data to train a convolutional neural network (U-Net). It is part of a research project I did during my Master's degree at Mines Paris - Université PSL, in Paris, France.
It contains code to generate synthetic data, the U-Net model using TensorFlow, as well as the logs and results of the research project.
I'm glad to discuss about this project with anyone interest in it. Feel free to send me a message at [email protected] for more details.