Person re-identification with YOLOv5s

YOLOv5 repo: https://github.com/ultralytics/yolov5
Re-identification model paper

Setup

1. Open Anaconda Prompt and navigate into project directory cd path_to_repo
2. Run conda env create from project directory (this will create a brand new conda environment).
3. Run conda activate reid_torch (if you want to run scripts from your console otherwise set the interpreter in your IDE)

• If you want, you can install packages from requirements.txt (pip install -r requirements.txt on terminal)

4. Download weights here. And place them in the paths: person_reid_yolo/reid/logs and person_reid_yolo/reid/model.

That's it! It should work out-of-the-box executing environment.yml file which deals with dependencies.

Getting Started

First of all, you need to place your video file in the directory: person_reid_yolo/test_data/video.mp4. And change the corresponding constant in the config file:

person_reid_yolo/config.py

Line 20 in c2d3ffe

SOURCE_VIDEO_FILE_PATH = f"{ROOTPATH}/test_data/we_1.mp4"

Now you can run the file run.py:

python run.py

Demonstration

Settings

You can play with the settings in the config file (config.py), changing them to suit your video.

Visualisation

You can visualize a database where all the cut out pictures will be sorted into separate folders for each person.

python create_db_visualization.py

All frames that you can observe after running run.py are saved in the folder for recordings (default is recordings). After marking the video you can create a new one by collecting all the frames with the file make_video.py:

python make_video.py