One validation test we want to run is to check that the joint distributions of galaxy properties are preserved by the generative model.
At zeroth order, I guess that means "correct" joint distributions of size, shape and flux, but also additional morphology statistics. I guess we could condition the image generation on some properties like redshift, and see if we recover all of these correctly.
One validation test we want to run is to check that the joint distributions of galaxy properties are preserved by the generative model.
At zeroth order, I guess that means "correct" joint distributions of size, shape and flux, but also additional morphology statistics. I guess we could condition the image generation on some properties like redshift, and see if we recover all of these correctly.