January 27, 2026
With the ample amount of transit clips we record and store from our livestream, it is natural to ultilize this data in a computer vision capacity. We are currently using a YOLO (You Only Look Once) model, specifically YOLO26s, trained on our custom dataset to detect and classify ships.
The current v1 dataset includes over 69,000 video frames and 3 object classes; Dry Bulk Cargo Ship, Tugboat, and Tug & Barge. With more to be hopefully added and re-trained soon. As we are a one man band it was not-practical to label all the videos manually, but thanks to Meta's SAM3 (Segment Anything Model) and Ultralytics we were able auto-annotate every frame. YOLO26s is able to run inference on Edge devices and we're using a NVIDIA Jetson Orin Nano 8GB.
