1 use the offical code to export object detection in concise format by: python3 label2det.py bdd100k/labels/bdd100k_labels_images_train.json \ bdd100k/detection_train.json 2 use python2 convert_to_coco_format.py as follows usage: convert_to_coco_format.py [-h] [--src SRC_PATH] [--dst DST_PATH] [--dir IM_DIR] convert the file to coco format optional arguments: -h, --help show this help message and exit --src SRC_PATH /path/to/source --dst DST_PATH /path/to/save.json --dir IM_DIR /path/to/image/