Manual object counting in images and videos is a time-consuming task. Counting the number of cells is an integral part of all cell-related research. A large amount of researches held on cells compare their results based on how many cells are present in analyses, thus there is a need for a good solution. This paper describes the use of a Convolutional Neural Network as a solution to the problem. CentroidNet and Mask R-CNN are proposed as possible solutions. As a result of the experiment, it is concluded that CentroidNet performs with higher accuracy than Mask R-CNN with relative scores of 88.41% and 14.86%. However, on the BloodCell dataset Mask R-CNN performs better than CentroidNet. By interpreting the results it is safe to say that CentroidNet is a good solution to the problem, but there is the need for a bigger and better-annotated dataset to test if that could help in gaining a higher performance in future projects.
Poster and paper not available.
Senne Root and Robert Slomp
