Abstract
This work addresses the problem of recognizing action categories in videos for which no training examples are available. The current state-of-the-art enables such a zero-shot recognition by learning universal mappings from videos to a shared semantic space, either trained on large-scale seen actions or on objects. While effective, we find that universal action and object mappings are biased to their seen categories. Such biases are further amplified due to biases between seen and unseen categories in the semantic space. The compounding biases result in many unseen action categories simply never being selected during inference, hampering zero-shot progress. We seek to address this limitation and introduce universal prototype transport for zero-shot action recognition. The main idea is to re-position the semantic prototypes of unseen actions through transduction, i.e. by using the distribution of the unlabelled test set. For universal action models, we first seek to find a hyperspherical optimal transport mapping from unseen action prototypes to the set of all projected test videos. We then define a target prototype for each unseen action as the weighted Fréchet mean over the transport couplings. Equipped with a target prototype, we propose to re-position unseen action prototypes along the geodesic spanned by the original and target prototypes, acting as a form of semantic regularization. For universal object models, we outline a variant that defines target prototypes based on an optimal transport between unseen action prototypes and semantic object prototypes. Empirically, we show that universal prototype transport diminishes the biased selection of unseen action prototypes and boosts both universal action and object models, resulting in state-of-the-art performance for zero-shot classification and spatio-temporal localization.
Abstract (translated)
URL
https://arxiv.org/abs/2203.03971
