TrackEval Lite

This is a simpler and more user-friendly implementation of the TrackEval script. It evaluates object tracking performance using various metrics.

Key Differences from the Original TrackEval

TrackEval Lite simplifies the usage and setup process compared to the original TrackEval repository. Here are the key differences:

• Simplified Input: Only requires ground truth and tracker result files.
• No Directory Structure Requirements: Does not require a specific directory structure or naming style.
• Direct Sequence Length Input: Instead of requiring a seqinfo.ini file, you directly input the sequence length.
• Ease of Use: Streamlined argument handling and setup, focusing on simplicity and ease of use.

Currently implemented metrics

Metric Family	Sub metrics	Paper	Code
HOTA metrics	HOTA, DetA, AssA, LocA, DetPr, DetRe, AssPr, AssRe	paper	code
CLEARMOT metrics	MOTA, MOTP, MT, ML, Frag, etc.	paper	code
Identity metrics	IDF1, IDP, IDR	paper	code
VACE metrics	ATA, SFDA	paper	code

Input Format

The script requires two files: a ground truth file and a tracker result file. Both files should follow this format:

<frame_id>,<track_id>,<bbox_left>,<bbox_top>,<bbox_width>,<bbox_height>

Example

1,1,992,212,57,50
2,1,1002,220,60,55
...

Installation

git clone https://github.com/30-A/trackeval_lite
cd trackeval_lite
pip install -r requirements.txt

How to Run

python scripts/run_mot_challenge.py \
  --GT_PATH path/to/gt.txt \
  --TRACKER_PATH path/to/tracker.txt \
  --SEQ_LENGTH 1000 \
  --METRICS HOTA CLEAR VACE Identity

Arguments

The script accepts the following command-line arguments:

• --GT_PATH (str): Path to the ground truth file.
• --TRACKER_PATH (str): Path to the tracker result file.
• --SEQ_LENGTH (int): Length of the sequence in frames.
• --METRICS (list): List of metrics to evaluate. Default: ['HOTA', 'CLEAR', 'Identity', 'VACE']
• --THRESHOLD (float): Threshold value for evaluation. Default: 0.5

Output

The script generates:

1. Evaluation Results: Metrics values printed in a tabular format.
2. HOTA Curve Plot: A plot of the HOTA metric over the sequence, saved as HOTA_curve.png in the current directory.

Timing analysis

TrackEval	MOTChallenge	Speedup vs MOTChallenge	py-motmetrics	Speedup vs py-motmetrics
9.64	66.23	6.87x	99.65	10.34x

Citing TrackEval

If you use this code in your research, please use the following BibTeX entry:

@misc{luiten2020trackeval,
  author =       {Jonathon Luiten, Arne Hoffhues},
  title =        {TrackEval},
  howpublished = {\url{https://github.com/JonathonLuiten/TrackEval}},
  year =         {2020}
}