This repository contains documentation and code for object detection. The primary software package we use is called darknet_ros, running on Ubuntu 18.04 with ROS Melodic.
Never push your changes directly to the main branch! Create a branch for each high level change you are working on (documentation or new-cv-algorithm or something). When you've finished working on and testing your code, open a pull request in Github, and describe the changes you've made and what needs to be reviewed by other team members. Don't forget to merge in any new changes from main into your branch. Once a couple of team members have reviewed and approved your changes, those changes will be merged into main! See this link for a quick guide on working with different git branches.
Follow the installation instructions for darknet_ros on the README. Make sure CUDA and OpenCV are installed. For CUDA, make sure you follow all the installation steps, including the mandatory post-installation steps. I've had some trouble with OpenCV 4, so I usually just install OpenCV 3.4.2 using this script. Download the script, change the OPENCV_VERSION to 3.4.2 and run the following command.
bash install-opencv.sh
You will also need to install CUDA following these instructions. Don't forget to follow the mandatory post-installation instructions!
If you run into any Nvidia or CUDA issues, try updating the graphics card drivers.
sudo apt update
sudo apt install nvidia-driver-{latest version number}
You'll need to reboot the computer after installing new graphics card drivers.
See the README in darknet_ros, but the basic gist is to change the subscribed topic in darknet_ros/config/ros.yaml to the topic that the camera node is publishing to, and change the cfg file to point to the right weights file.
Use DarkMark (need to have darknet installed) or labelImg (works on any OS) to annotate images, then follow the steps here to start the training. We'll most likely be using yolo-v3-tiny, so make sure you use the right starting weights and cfg files. Can be done with the darknet submodule within the darknet_ros package.
idk...