Wyhsirius

lia

The model presented in this work aims to learn a representation of images such that interpolation in the latent space leads to smooth and realistic animation. The authors propose the use of a variational autoencoder (VAE) as the backbone of their model, with an additional conditioning network to guide the animation process. They introduce a new dataset, called KittiCars, for training and evaluation. The results demonstrate that the model is capable of generating visually appealing animations with realistic motion.

Read more