Master thesis : Cell Counting

Abstract

Counting the number of objects inside a given image is a trivial task for most of us but can become daunting and error prone when the number of objects to count grows larger. As counting has a wide variety of use cases such as crowd monitoring, traffic management, wild life census or even medical diagnosis, having an automated method to count these objects would simplify the work of many. This work focuses on the task of counting cells inside a given image. While an extensive amount of work have proposed new methods to automate this tasks or new annotated datasets to improve training, there has been a lack of benchmark over the quality of these methods. This work proposes to solve this problem by performing a thorough review of the current literature and by benchmarking the different methods using the same standard way. The most recent models have been implemented using the same framework and trained in the same fashion. The goal of this work is, we hope, to provide an accurate assessment of what the current methods can do, which of them are the better and in which situation and where there be further improvements in this area of research.