Updates 3/2/24

For the last few months, we've been working on our vision system. We're currently evaluating the pros and cons of both SSD (single shot detection) and YOLO models. We've implemented a custom implementation of SSD and are tinkering with Ultralytic's YOLOv8 model. Here are our current thoughts on both models. YOLOV8: Very fast and easy to implement, fast training times, has trouble detecting smaller weeds SSD: Fast performance, difficult to implement, slow training times, high accuracy, needs lots of data

We've also created custom software for labeling and augmenting data. We created FastYOLO as a way for people to annotate and augment image data locally and without having to pay for the expensive online SaaS providers such as Roboflow. FastYOLO was created during the 8th annual Los Altos Hackathon and won the award for best solo hack.