Going from a Vector (attributes) to a Matrix (Image), Reverse CNN?

Do any of you have an example showing the inputs of a NN being a vector and the outputs being a matrix? I would like to be able to input multiple inputs lets say facial features (race,sex,hair color, hair length, eye color, eye spacing, nose length, nose width, mouth width, mouth depth) and output an image of a generic/average face that represents those terms. I would appreciate you all's help because I am having major issues converting the ML speak into useful information.

Answer 1

[Walter Roberson]

This is done all the time, such as for facial recognition. The features are generated for all of the representative images and (possibly) stored. Then the features are generated for the input image. The "distance" between the features for the input and the features for the stored images is generated, and the few closest in feature-space are considered to be likely matches.

In your situation, it would not be necessary to generate the probe features from an input image: you could construct that input however you wanted.

The role that Machine Learning can play in all of this is to help prune down which features are most important for a high recognition rate.

[Blake Van Winkle]

To Walter:

I understand that it can be done, but I haven't seen any example of this done in Matlab land. Do you have such of an example.

Thanks,