A (back and forth) trip down latent space


With the metaphor of latent space, we are asked to conceive the trained network as a system of multidimensional coordinates that we could walk. If we can create this digital space as a consequence of the training of a GAN, what happens if it arises from a specific space and route? What is the result of duplicating a real space in a latent space? How is it transited and how it is transformed, this original space, when we position ourselves in the virtual space of the neural network?

A trip down Market Street is a 1906 movie that travels through the street of the same name in San Francisco. It is an example of a particular genre of early cinema called phantom ride: filming a route from a mobile point of view (a tram in this case) both for the attraction that is the space in itself as for the kinetic effect of the mobile point of view. All the frames of this movie constituted the training dataset for the network.