Interpolation between captions

[philippmoehl]

Hi, I built an interpolation between two different noise and local noise vectors for the model. Now I want to do the interpolation between captions. Any ideas how this could work out?

For the sentence embeddings this should go through. But the caption is also converted to a l tensor of word embeddings with length of the caption. This is used in the second and third stage of the generator and sets the shape of the hidden values given to the next stage and to GET_IMAGE_G. From my point of view the normal way of interpolation like with Z vectors is not possible here.