Context This came out of [the work of Mat Kelcey][1] where he used a fully convolutional network to count bees. I am very interested in learning about this approach and also how to label data with semi-supervised learning. Content The initial dataset had 100 labelled images. I have now supplemented this with 500 labelled images and 7000 approx unlabelled images. The images were taken at close range and top down as I was hoping to see if parasites such as varroa could be seen. I will be adding different angles and fields as I gather them with my hacked together picam Acknowledgements Mat Kelcey for coming up with the idea Xiaofan for all her help getting me to understand the intricacies of CNNs Inspiration Whats all this semi supervised learning about? Can you really bootstrap an algorithm with a couple of images and use it to labels its own data? [1]: http://matpalm.com/blog/counting_bees/