Multi-Object Tracking Example

Introduction

Multi-Object Tracking (MOT) technology is used to simultaneously identify and track multiple targets within video sequences, involving the association of targets across different frames.

Usage

X-AnyLabeling is enhanced with a versatile tracking system capable of handling multiple tasks:

• Object Detection
• Oriented Bounding Boxes Object Detection
• Instance Segmentation
• Pose Estimation

It incorporates advanced tracking algorithms such as TrackTrack, ByteTrack and Bot-Sort to ensure robust and accurate tracking across these diverse tasks.

Here's how to perform multi-object tracking using the tool:

1. Load the video file, e.g., Bangkok.mp4. (Note: The file path must not contain Chinese characters!)
2. Load a built-in tracking model from the table below, or load a custom tracking model.
3. Click to run, and after verifying that everything is correct, you can use the shortcut Ctrl+M to run all frames at once.

Built-in tracking model configurations:

Task	Tracker	Model config
Object Detection	BoT-SORT	yolov5s_det_botsort, yolov8s_det_botsort, yolo11s_det_botsort
Object Detection	TrackTrack	yolo26s_det_tracktrack
Oriented Bounding Box Detection	BoT-SORT	yolov8n_obb_botsort, yolo11s_obb_botsort
Oriented Bounding Box Detection	TrackTrack	yolo26s_obb_tracktrack
Instance Segmentation	ByteTrack	yolov8m_seg_bytetrack
Instance Segmentation	BoT-SORT	yolo11s_seg_botsort
Instance Segmentation	TrackTrack	yolo26s_seg_tracktrack
Pose Estimation	BoT-SORT	yolov8x_pose_p6_botsort, yolo11s_pose_botsort
Pose Estimation	TrackTrack	yolo26s_pose_tracktrack

In this process, the group_id field represents the track_id of the current target box. During actual tracking, the first few frames may not have a group_id if the object confidence score is below the tracking threshold.

Export

For instructions on exporting MOT annotations, please consult the user guide available:

• English version
• Chinese version