Learning the Predictability of the Future
This work addresses the problem of learning what is predictable in future video frames for computer vision researchers, offering an incremental improvement in handling uncertainty in prediction.
This paper introduces a framework that learns from unlabeled video to identify predictable features in the future, rather than pre-committing to specific features. It proposes a predictive model in hyperbolic space that automatically adjusts its prediction granularity based on confidence, predicting at a concrete level when confident and a higher level of abstraction when not.
We introduce a framework for learning from unlabeled video what is predictable in the future. Instead of committing up front to features to predict, our approach learns from data which features are predictable. Based on the observation that hyperbolic geometry naturally and compactly encodes hierarchical structure, we propose a predictive model in hyperbolic space. When the model is most confident, it will predict at a concrete level of the hierarchy, but when the model is not confident, it learns to automatically select a higher level of abstraction. Experiments on two established datasets show the key role of hierarchical representations for action prediction. Although our representation is trained with unlabeled video, visualizations show that action hierarchies emerge in the representation.