flavour tagging, the identification of jets containing b and c hadrons, is key to many fundamental measurements and searches for new phenomena performed by the ATLAS experiment. the existing algorithms for flavour tagging in ATLAS use a few different tools to identify the distinctive topology of b->c decay which gives rise to two secondary vertices with high invariant mass and several associated charged tracks. Machine learning is used to combine the information from these tools to infer the probability of a jet to be a b,c or light jet. We have created 2D images of the charged tracks associated to the jets, and trained a convolutional neural network (CNN) with these images as input. I will discuss the possible ways this information is used for flavour tagging, the types of neural net architectures, and compare the resulting performance to existing flavour tagging algorithms.