PromptTrack

📄 Overview

PromptTracker is a powerful object tracking tool that enables the detection and tracking of objects in videos using text-based prompts. It supports advanced object detection models such as MDETR and OWL-VITV2, providing robust tracking with options to handle fixed populations in enclosed environments.

📥 Installation

To install PromptTracker, use the following command:

pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ PromptTrack (version test)

pip install --no-deps bytetracker

The package has been implemented for python from 3.9 to earlier (last version currently 3.12)

You can check our github repo

Usage

Step 1 initialization

from PromptTrack import PromptTracker

#initialize PromptTrack

tracker = PromptTracker()

Step 2: Object Detection

Detect objects in a video using a text prompt:

Parameters:

• video_file (str): Path to the input video file.

• prompt (str): Comma-separated names of entities to track (e.g., "pig,bird").

• nms_threshold (float): Non-Maximum Suppression threshold for filtering overlapping detections.

• detection_threshold (float): Confidence threshold for reporting detections.

• detector (str): Detection model to use ("MDETR" or "OWL-VITV2").

tracker.detect_objects(video_file, prompt="penguin",nms_threshold=0.8, detection_threshold=0.3 ,detector="OWL-VITV2")

Step 3: Tracking Objects

Process the video using a Multiple Object Tracker (MOT):

Parameters:

• video_file (str): Path to the input video file.

• fixed_parc (bool, default=True): Set to True if the number of tracked objects remains constant.

• track_thresh (float, default=0.40): Detection confidence threshold for the tracker.

• match_thresh (float, default=1.0): Matching threshold between new detections and existing tracks (1 to match all, 0 to match none).

• frame_rate (int, default=6): Frame rate of the video.

• max_time_lost (float, default=inf): Maximum time an object can be missing before being removed.

• nbr_frames_fixing (int, default=300): Number of frames considered for tracking in a fixed setting.

tracker.process_mot (video_file, fixed_parc=True,track_thresh=0.40, match_thresh=0.8, frame_rate=25,max_time_lost=float('inf'),nbr_frames_fixing=800)

Step 4: Read Processed Video

To visualize the tracked objects in the processed video:

Parameters:

• video_file (str): Path to the input video file.

• fps (int, default=20): Frames per second for video playback.

tracker.read_video_with_mot(video_file,fps=20)

Full example usage

tracker = PromptTracker()



video_file = "[path_to_your_video]"

tracker.detect_objects(video_file, prompt="penguin", nms_threshold=0.8, detection_threshold=0.3, detector="OWL-VITV2")



tracker.process_mot(video_file, fixed_parc=True, track_thresh=0.40, match_thresh=1, frame_rate=25, max_time_lost=float('inf'), nbr_frames_fixing=800)



tracker.read_video_with_mot(video_file, fps=20)

Examples (on our github)