We propose MHM-GNN, an inductive unsupervised graph representation approach that combines jointk-node representations with energy-based models (hypergraph Markov networks) and GNNs.

Conditional GANs(cGAN), intheirrudimentary form,sufferfromcriticaldrawbacks such as the lack of diversity in generated outputs and distortion between the latent and output manifolds.

Text spotting has received increasing attention due to its various applications, such as video retrieval[1]andautonomousdriving[2].

Imagine trying to track one particular fruitfly in a swarm of hundreds. Higher biological visual systems have evolved to track moving objects by relying on boththeirappearance andtheirmotiontrajectories.