Forecasting Characteristic 3D Poses of Human Actions
This work addresses the problem of predicting semantically meaningful future human poses for applications requiring understanding of human intent, such as human-robot interaction or animation, by decoupling pose prediction from fixed time intervals.
This paper introduces the task of forecasting characteristic 3D poses, predicting a future action-defining pose from a short observation sequence, rather than fixed-interval frame-by-frame prediction. Their probabilistic approach, which models multi-modality and samples future poses autoregressively, outperforms state-of-the-art methods by 26% on average.
We propose the task of forecasting characteristic 3d poses: from a short sequence observation of a person, predict a future 3d pose of that person in a likely action-defining, characteristic pose -- for instance, from observing a person picking up an apple, predict the pose of the person eating the apple. Prior work on human motion prediction estimates future poses at fixed time intervals. Although easy to define, this frame-by-frame formulation confounds temporal and intentional aspects of human action. Instead, we define a semantically meaningful pose prediction task that decouples the predicted pose from time, taking inspiration from goal-directed behavior. To predict characteristic poses, we propose a probabilistic approach that models the possible multi-modality in the distribution of likely characteristic poses. We then sample future pose hypotheses from the predicted distribution in an autoregressive fashion to model dependencies between joints. To evaluate our method, we construct a dataset of manually annotated characteristic 3d poses. Our experiments with this dataset suggest that our proposed probabilistic approach outperforms state-of-the-art methods by 26% on average.