Video Object Tracking report generator

Copyright (c) 2020-2021 Antmicro

A python tool for generating reports regarding detection and tracking of a single object in video sequences.

Check how this repository can be used in ros-tracking-policy-examples repository.

Usage

To generate reports run main.py script. It takes exactly two positonal arguments: output_path and input_path, where:

• output_path is path to directory with test results; format of this repository is specified in the ros-tracking-nodes-tester-node repository.
• input_path path to directory containing dataset which was used to run the tests; the format is also described in the above-mentioned repository.

Generated reports

Reports are generated as html files and .png images.

The index.html is the root of the report. It contains a list of tested tracking policies.

For each policy, there is a summary and the performance per sequence.

The summary provide:

• Mean IoU for each video sequence.
• Plot describing the mean IoU depending on the size of the ground truth bounding box size.
• Mean, median and standard deviation for the IoU for all sequences.

The performance per sequence shows:

• Processed frames per second by the policy (mean, median, standard deviation),
• Mean, median and standard deviation for IoU metric,
• IoU changes over time for a given video sequence (this can tell which parts of the video sequence were especially tough for the policy),
• Inference time for each frame.