The goal of this project is to analyze how noisy labels affect a model’s performance by segmenting and classifying biomedical images. We will use deep learning models in Pytorch and a standard biomedical imaging dataset (MoNuSAC). The final outcome is a deep study of the performance of a model that segments and classifies a multi-class dataset with and without imperfect annotations. This analysis will allow us to know how the algorithm behaves with noisy datasets and find suitable strategies to deal with each type of imperfection.
Required skills: Programming in Python, knowledge of deep learning libraries and computer vision technologies is preferred.
Contacts: christine.decaestecker@ulb.be, alberto.franzin@ulb.be