Adopt a labeling pipeline

[nickswalker]

Labeling images is a big part of @Home (witness: this Homer blog post).

Now that we're starting to see detections out of off-the-shelf solutions in #5, we'll soon turn towards fine tuning with in-domain data. We need a tool that:

• Lets human annotators quickly mark bounding boxes or segmentation masks and
• Produces standard annotation formats

What other do:

• Villa homerolled one in Python ontop of OpenCV's gui stuff and grabcut. It didn't produce actual segmentation masks, but would rather use the segmentation to cut out the object and put in random backgrounds as data augmentation. It would dump out bbox annotations for training YOLO
• TUe's seems to be C++ on top of OpenCV gui

We really shouldn't have to build something this generic. Searching GitHub, I can already see some popular options
https://github.com/wkentaro/labelme
https://github.com/abreheret/PixelAnnotationTool
https://github.com/kyamagu/js-segment-annotator
https://github.com/hanskrupakar/COCO-Style-Dataset-Generator-GUI

We should find the most-used option that addresses our needs

[nickswalker]

This one looks promising: https://github.com/opencv/cvat