Jul 1, 2020 – 11:00 AM
In general sketches are generated using generative adversarial models. We propose a new approach which represents the sketches as sequences of segments. This means that segments are ordered with respect to time, so we organize them considering a predefined criterion. We train an RNN using such sequences and then we use the RNN to generate new sketches. After training we provide the network some initial segments and let it complete the sketch. In this approach the sketch generation occurs as in natural language processing where the next word is predicted.