Cardiac MRI segmentation is a clinically interesting field that can accelerate and improve diagnostics. Targeting the capability of models towards better generalizing in unseen subsets of data that can better represent minority cohorts, greatly enhancing the lives of multiple people, cheapening the diagnostics, and making current models more resilient to unseen pathologies. In this project, our aim was to study how different architectures behave in a multiview multivendor multipathology scenario with respect to these generalization capacities and explore how postprocessing can improve the results. In addition, we also assess the computational cost that these models need to ensure that they are valid for clinical products and machines that can be reached at any clinical center.