Perceptual difference image

[EstebanMontandon]

Hello guys
I'm trying to compute the perceptual difference between two images. What I'm trying to achieve is something similar to a perceptual difference map of the same dimension as the input images. Similar to the perceptual difference referenced in this paper "Generalizing Unsupervised Anomaly Detection: Towards Unbiased Pathology Screening, Cosmin I. Bercea.".

I've tried using the class PerceptualLoss (from generative/losses/), but this only returns a perceptual similarity loss score.

Is there any implementation example that generates this perceptual difference? thanks in advance.
regards