Generation of patient specific cardiac chamber models using generative neural networks under a Bayesian framework for electroanatomical mapping