This version is another training experiment I've been wanting to try. I cropped the dataset images and rotated most of them sideways or upside down, then I randomly shuffled the tags between the images before training so the AI would be looking for things that weren't always there.