This project is still under development.
In this notebook, I’ll perform a full implementation of YOLOv3 in PyTorch based on the following materials:
- Orginial implementation of darknet
- YOLOv3: An Incremental Improvement, Joseph Redmon, Ali Farhadi
- How to implement a YOLO (v3) object detector from scratch in PyTorch, Ayoosh Kathuria.
Example of image detection output:
Project was created with Python 3.6.x and pyTorch 0.4.0.
Clone repository and install requirements:
git clone https://github.com/BarbaraStempien/DL--YOLO-Real-Time-Object-Detection.git
cd DL--YOLO-Real-Time-Object-Detection
pip3 install -r requirements.txt
Download YOLOv3 weights:
wget https://pjreddie.com/media/files/yolov3.weights
You can download different weights from the YOLO website.
python image_detection.py -h
-h allows you to look up all terminal flags
python image_detection.py --images imgs --det det
--images flag defines the directory of your input images
--det flag defines the directory of your output imanges (directory to which images with detected objects will be saved)
python image_detection.py --images imgs --det det --reso 320
--reso flag allows you to change resolution of the input images, default value is 416. Number must be a multiple of 32 and greater than 32.
python image_detection.py --confidence 0.25
--confidence flag allows you to change confidence level. By default, model displays object detected with a confidence of 0.5 or higher.
python video_detection.py -h
-h flag allows you to look up all terminal flags defined.
python video_detection.py --video video/example.avi
--video flag defines the directory and name of your input video. Video must be in the the .avi format.
python video_detection.py
Run video detection script without a --video flag to use machine's default camera.
python video_detection.py --reso 320
--reso flag allows you to change resolution of the video, default value is 160.
python video_detection.py --confidence 0.25
--confidence flag allows you to change confidence level. By default, model displays object detected with a confidence of 0.5 or higher.
To stop web cam recording, press Q on the keyboard.
If you want to contribute, check out CONTRIBUTING.md.