Abstract:
There is an ever-growing need for automated detection (and classification) of
microscopic images containing cellular samples. To this end, the focus of this work is
to provide a method of performing this detection: through the implementation of a
Faster RCNN model. The .tiff images for training and testing are fed to the network,
with various hyperparameter adjustments between runs, and the anchors/bounding
boxes are calculated by FRCNN. Four separate output loss functions are calculated
and then unified for a final metric. The network utilized an underlying VGG-16
architecture, and a RPN (Region Proposal Network) which is responsible for the
aforementioned bounding boxes. The architecture is run on keras (tensorflow
backend).
