This package is for creating, training, and applying a convolutional neural network defined using the yolo model framework (Redmand et al.). The hallmark feature of this model framework is that input images can have any number of channels, and the network will create a representation of the input image containg exactly 3 channels. This simplifies downstream operations and allows the user to apply this model to images with a variable number of channels.
There are 3 major steps in this pipeline: (1) Preprocess the input image (2) Apply the model to the pre-processed image (3) Postprocess the output
Normalize the pixel/voxel values with respect to all these values in a given image.
(Optional) Apply a gamma correction (gamma = 0.5) to the input image, to adjust for abnormally large variations in brightness across the image
(Optional) Upsample the input image by a factor of 2. Useful if target objects only occupy a few pixels (i.e. cell nuclei segmentation)