skuzminoff / multi-person-tracker

Simple Multi Person Tracker implementation in PyTorch

Simple Multi Person Tracker

Simple and easy to use multi person tracker implementation. This project supports YOLOV3 & MaskRCNN as detector and SORT as object tracker.

Installation

First you need to install the requirements:

$ pip install -r requirements.txt

Then install the package via:

$ pip install git+https://github.com/mkocabas/multi-person-tracker.git

Usage

Run examples/demo_video.py for a minimally working example. Here is a sample:

from multi_person_tracker import MPT
from multi_person_tracker.data import video_to_images

image_folder = video_to_images('sample_video.mp4')

mpt = MPT(
    display=True,
    detector_type='yolo', # or 'maskrcnn'
    batch_size=10,
    yolo_img_size=416,
)

result = mpt(image_folder, output_file='sample_output.mp4')

Runtime Performance

Detector	Tracker	GPU	FPS
MaskRCNN	Sort	RTX2080Ti	13
YOLOv3-256	Sort	RTX2080Ti	120
YOLOv3-416	Sort	RTX2080Ti	80
YOLOv3-608	Sort	RTX2080Ti	46

Important Note

• Install torchvision from the source as done in requirements.txt to be able to use the best performing MaskRCNN pretrained model. Check this Issue and PR for more details.

References

• YOLOV3: paper
• MaskRCNN: paper
• SORT: paper