Understanding Object Dynamics for Interactive Image-to-Video Synthesis
This work addresses the challenge of interactive image-to-video synthesis for applications in graphics and AI, offering a novel method for user-controlled deformation prediction, though it builds incrementally on existing video prediction techniques.
The paper tackles the problem of predicting how objects deform over time in response to local user interactions, such as poking a pixel, by learning object dynamics from videos without explicit manipulation data. The result is a generative model that enables interactive control of deformations and transfers dynamics to novel objects, outperforming common video prediction frameworks in experiments.
What would be the effect of locally poking a static scene? We present an approach that learns naturally-looking global articulations caused by a local manipulation at a pixel level. Training requires only videos of moving objects but no information of the underlying manipulation of the physical scene. Our generative model learns to infer natural object dynamics as a response to user interaction and learns about the interrelations between different object body regions. Given a static image of an object and a local poking of a pixel, the approach then predicts how the object would deform over time. In contrast to existing work on video prediction, we do not synthesize arbitrary realistic videos but enable local interactive control of the deformation. Our model is not restricted to particular object categories and can transfer dynamics onto novel unseen object instances. Extensive experiments on diverse objects demonstrate the effectiveness of our approach compared to common video prediction frameworks. Project page is available at https://bit.ly/3cxfA2L .