dataset information

detector used: YOLOv3 trained on COCO.

Detections have been manually associated to categorize which detections belong to the same object.

detection format:

'object_key': integer representing the unique object id

'image_key': integer representing image key

'image_path': relative path in the TUM dataset

'box': float image coordinates in xmin,ymin,xmax,ymax format

'objectness': float objectness score

'scores': softmax class scores (80)