Check in for Nov 30, 2021
Introduction:
We propose DeepAnimation, a vector-based generation model that predicts the next possible keyframes for animation given a SVG input in real-time. The goal is to help users create web-friendly, scalable, and lightweight animations by recommending exemplars from existing commercial-grade ones about “where”, “how”, or “when” should the animation occur. This model helps users unfamiliar with underlying parametric concepts in creating vectorized images to swiftly iterate on animations. It can be used to fabricate web posters and other graphical elements or to enrich the virtual conversation experience. We additionally present a GUI demonstrating how recommendations can be made as the user is drawing.
Challenges: What has been the hardest part of the project you’ve encountered so far?
The hardest part of the project is how do we control the latent vector and to get meaningful animation generation. If we can interpret the direction of change, such as those used in age editing, we would be able to control the latent vector in our favor. Currently, our model pipeline takes in a user-created SVG file on our web-based interface (see the image below). Once a user creates a drawing, it is transformed into a point in high dimensional space. Then, we generate a circular path that goes through this high dimensional point. This circular path serves as an agent to create looping animations. Finally, the decoder would take a fixed number (N) of points uniformly sampled from the high dimensional circle to generate a sequential image as the final result.
Insights: Are there any concrete results you can show at this point?
We pre-trained a MNIST network that expands the variational autoencoder base code used in our assignment. Preliminary results show that transitions between the animations are smooth, indicating that the hyper dimensional circle provides a recognizable, sequential pattern from the user inputs. Here is the link of the suggested animation from a user sketching a number “5” (Result showing transition from a user-input sketch of “5” to automatically generated “3”)
How is your model performing compared with expectations?
For our generative model, we rely on visualization and human judgement for comparison. The model fits our expectation in terms of transition and encoder-decoder for customized input. However, the model still lacks the ability to consider the relationship between the input shape and the output animation at this stage.
Are you on track with your project?
We are a bit lagged behind the progress since there is not so much preliminary work in this area. We are trying to make great innovations in this area.
What do you need to dedicate more time to? What are you thinking of changing, if anything?
We need to spend more time on the core model development in terms of inferring animation from the input svg.
Log in or sign up for Devpost to join the conversation.