Cell detection

In this project we made a fully-convolutional network (FCN) for the detection of blood cells in microscopy, see Figure 2. The training of the FCN was performed on data-augmented image patches extracted from microscopy images. Every epoch, new training data was used. To achieve a better detector, the training data was expanded by augmentation (affine transformations), hard examples were taken into account and sub-pixel precision was used. In the end, a cell detector with good precision values was created.