Hamiltonian latent operators for content and motion disentanglement in image sequences

Neural Information Processing Systems 

We introduce HALO - a deep generative model utilising HAmiltonian Latent Operators to reliably disentangle content and motion information in image sequences.