Work

Exploring a method to evaluate image-conditioned deep generative models for their capacity to reproduce domain-relevant spatial context