AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions

TLDR

Existing spatio‑temporal action datasets provide sparse annotations for composite actions in short clips, whereas AVA offers dense, atomic action labels over longer videos. This paper introduces the AVA dataset of spatio‑temporally localized atomic visual actions and proposes a benchmark method that outperforms prior models on JHMDB and UCF101‑24. AVA contains 80 atomic actions annotated in 437 fifteen‑minute movie clips, yielding 1.59 million spatio‑temporal labels, with multiple labels per person, exhaustive coverage, temporal linking across segments, and precise localization. Although the benchmark achieves state‑of‑the‑art performance on existing datasets, its mAP on AVA is only 15.8 %, highlighting the dataset’s difficulty and the need for new video‑understanding approaches.

Abstract

This paper introduces a video dataset of spatio-temporally localized Atomic Visual Actions (AVA). The AVA dataset densely annotates 80 atomic visual actions in 437 15-minute video clips, where actions are localized in space and time, resulting in 1.59M action labels with multiple labels per person occurring frequently. The key characteristics of our dataset are: (1) the definition of atomic visual actions, rather than composite actions; (2) precise spatio-temporal annotations with possibly multiple annotations for each person; (3) exhaustive annotation of these atomic actions over 15-minute video clips; (4) people temporally linked across consecutive segments; and (5) using movies to gather a varied set of action representations. This departs from existing datasets for spatio-temporal action recognition, which typically provide sparse annotations for composite actions in short video clips. AVA, with its realistic scene and action complexity, exposes the intrinsic difficulty of action recognition. To benchmark this, we present a novel approach for action localization that builds upon the current state-of-the-art methods, and demonstrates better performance on JHMDB and UCF101-24 categories. While setting a new state of the art on existing datasets, the overall results on AVA are low at 15.8% mAP, underscoring the need for developing new approaches for video understanding.