The overlay.py script would overlay object images on the background image, and create labels. The script will attempt to create directories train, test, and labels, and creates a file sphero_label_map.pbtxt.
To run it:
python overlay.py 100
TensorFlow Object Detection API requires a specific format of the input data. The convert_to_tfrecord.py script takes care of that, after images have been generated.
To create train.record:
python convert_to_tfrecord.py --output_path train.record --images_dir train --labels_dir labels