GenHowTo: Learning to Generate Actions and State Transformations from Instructional Videos
This work addresses the challenge of creating realistic visual content for instructional or simulation purposes, representing a strong specific gain in image generation.
The paper tackles the problem of generating images that depict actions and object state transformations from an input image and text prompt, achieving 88% and 74% accuracy on seen and unseen interaction categories, respectively.
We address the task of generating temporally consistent and physically plausible images of actions and object state transformations. Given an input image and a text prompt describing the targeted transformation, our generated images preserve the environment and transform objects in the initial image. Our contributions are threefold. First, we leverage a large body of instructional videos and automatically mine a dataset of triplets of consecutive frames corresponding to initial object states, actions, and resulting object transformations. Second, equipped with this data, we develop and train a conditioned diffusion model dubbed GenHowTo. Third, we evaluate GenHowTo on a variety of objects and actions and show superior performance compared to existing methods. In particular, we introduce a quantitative evaluation where GenHowTo achieves 88% and 74% on seen and unseen interaction categories, respectively, outperforming prior work by a large margin.