Deep Learning Training Data Formats

[bkaplan1]

I'm interested in using the DL training data available at Source Cooperative.   The format of these training data is 

The source files are tif files and the label files are geojson files.  However, the metadata format (I guess these means the labels) of the exported data must be Classified_Tiles, PASCAL_VOC_rectangles, or KITTI_rectangles.  Is there a way to convert the geojson files to the required metadata format.   I found GitHub - umautobots/vod-converter: Convert between visual object detection datasets to convert between KITTI and VOCs but not geojson.