Each dataset has an annotations.txt and an images dir.  The format of the annotations.txt is as follows:

ImageLabelManager
Version 2

<image filename>
<number of labels>
<label>
<label>
...

<image filename>
<number of labels>
<label>
<label>
...


and so on, where the label format is:

<track id> <class_name> <x coord of upper left hand point> <y coord of upper left hand point> <width> <height>

The origin of the image is at the upper left hand corner; x points right, and y points down.