Constraining neural networks output by an interpolating loss function with region priors

NeurIPS 2020