Our team member Leon Sixt has developed a new image processing component for the decoding of bee markers. The method is based on convolutional neural networks, the current state of the art in computer vision. ConvNets are awesome, however, they need large amounts of labeled data. In our case, this would have meant to click not only the image location of a tag but also the bit configuration for many thousand tag instances. Leons new solution “RenderGAN” (paper in prep) was to train a special network that can learn to reproduce realistic tag images. He then trained a decoder network using only generated tag images – no human labeling necessary!