implement a simple application pipeline

[abduld]

the pipeline would use instance segmentation to detect bounding boxes within the image, it would then crop the image to the specified coordinates and run image classification on each subimage