CLEAR-MOT METRICS in Python

Implementation of the multiple object tracking metrics as defined in this article:

Keni, Bernardin, and Stiefelhagen Rainer. "Evaluating multiple object tracking performance: the CLEAR MOT metrics." EURASIP Journal on Image and Video Processing 2008 (2008)

Link to article

Input data type

XML

XML file as the example below:

<?xml version="1.0" ?>
<Video end_frame="400" fname="/path/to/filename.jpg" start_frame="1">
	<Trajectory end_frame="400" obj_id="0" obj_type="car" start_frame="1">
		<Frame annotation="0" contour_pt="0" frame_no="1" height="91" observation="0" width="141" x="489" y="323"/>
		<Frame annotation="0" contour_pt="0" frame_no="2" height="95" observation="0" width="146" x="493" y="323"/>
		<Frame annotation="0" contour_pt="0" frame_no="3" height="96" observation="0" width="147" x="498" y="326"/>
	</Trajectory>
	<Trajectory end_frame="400" obj_id="4" obj_type="car" start_frame="1">
		<Frame annotation="0" contour_pt="0" frame_no="1" height="37" observation="0" width="40" x="272" y="195"/>
		<Frame annotation="0" contour_pt="0" frame_no="2" height="41" observation="0" width="23" x="125" y="303"/>
		<Frame annotation="0" contour_pt="0" frame_no="3" height="33" observation="0" width="23" x="623" y="197"/>
	</Trajectory>
	...
</Video>

Only some attributes are necessary for each tag:

Video

• end_frame
• start_frame

Trajectory

• obj_id

Frame

• frame_no
• width
• height
• x
• y

SQLITE (Polytrack)

SQLite table named "bounding_boxes" organized as follow (all entries in this table):

object_id	frame_number	x_top_left	y_top_left	x_bottom_right	y_bottom_right
1	200	22	76	194	211
2	200	400	387	453	444
...	...	...	...	...	...

Dependencies

1. Python 3
2. Munkres
3. Numpy

Usage

python3 MOT_Metrics.py path/to/annotations/file path/to/hypotheses/file evaluation_method distance

evaluation_method is either:

• bboverlap
• centroid

e.g.

python3 MOT_Metrics.py ./rene_gt.sqlite ./rene_vibe.sqlite bboverlap 0.5

License

See the LICENSE file for more details.