Our traffic light detector is only trained on simulator images and needs to be retrained on real-life images of traffic lights. But before that, we can simply test it out on real-images to see how far behind we are. Check [here] (https://github.com/subhash/CarND-Capstone#real-world-testing)
Once we have this testing setup, it'll be helpful if and when we retrain based on real images (I can help with this)
Run ROS system (roslaunch launch/styx.launch) and run ROS bag (rosbag play <xyz.bag>) to extract images under tl_detector (First, create a dir called image_dump under tl_detector)
Annotate images and retrain object detection model
Our traffic light detector is only trained on simulator images and needs to be retrained on real-life images of traffic lights. But before that, we can simply test it out on real-images to see how far behind we are. Check [here] (https://github.com/subhash/CarND-Capstone#real-world-testing)
Once we have this testing setup, it'll be helpful if and when we retrain based on real images (I can help with this)