Problems about Paper [ END-TO-END OPTIMIZED IMAGE COMPRESSION ]

[achel-x]

Recently, I read the seriese of papers that how to end to end compress image.

In this picture, input image x transforms to latent representation y by analysis transform. My question is concerned with z, how x transforms to z, or what is the g_p transform. I found the description in paper named it as perceptual transform, but I can't understand what exactly it is. And what is the relationship between z and y ?
I am not good at mathmatics, so it's tough for me to understand it from the formulas in your papers.

[jonaballe]

Hi,

in that paper z corresponds to a perceptual space, in which distances correlate better with perceived distances between images. For instance, this could be a colorspace such as YCbCr, or a more general perceptual space such as NLP or PIM.

Hope this helps!