Sugey is a home based AI platform for disabilities like Alzheimer's and autism. We use unique camera systems and a private in-home server to help families and individuals live better lives.
We're approaching releasing our MVP, and if you're interested in helping, be sure to contact us.
It takes anyone a few days at most to do computer vision with PJReddie's YOLO. You can plug it in, modify slightly, and go.
However, to really understand what's going on, and to be able to modify it for your own purposes, well, you should build it from scratch in your framework of choice.
So, we chose Pytorch, that part was easy. The hard part was the breakdown. A big thanks to Ayoosh Kathuria, who did a 5 part series on Paperspace walking through all of this code. We adjusted a few things here or there, but the bulk of the code comes from his work.
However, Ayoosh assumed some knowledge, and we needed to understand every piece. So, we added commenting to nearly every line of code explaining what was going on, which in conjunction with the tutorial, should make this a unique repo for YOLOv3.
You'll need to download the weights.
wget https://pjreddie.com/media/files/yolov3.weights
We highly suggest using a virtual environment of your choice, and then (assuming you setup your environment with Python 3) installing the requirements file:
pip install requirements.txt
To run the single image detector, in your virtual environment you can run:
python detect.py --images images/dog-cycle-car.png --det det
To run it on your webcam, you can run:
python demo.py
PJReddie put YOLO in the public domain. You can do whatever you want with it as long as you assume "AS IS", that means, he (and we) are not reponsible for how you use it. We're throwing the MIT license on just because of all the commenting and educational work involved. You can do whatever you want with this software under that MIT license.
We definitely take pull requests!
Our request is that anything you do, you comment well. An example in the comments is helpful, but we don't want 3 paragraphs. Try a happy balance.