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

TL;DR

AVA addresses the need for fine-grained, spatio-temporally localized action understanding by introducing 1 Hz annotations over 430 diverse 15-minute movie clips for 80 atomic actions, plus dense per-person labels and temporal linking. The dataset enables training and evaluation of new action localization methods, demonstrated by a baseline that fuses Faster R-CNN region proposals with 3D I3D tubes, achieving state-of-the-art on JHMDB/UCF101-24 but only modest AVA performance, underscoring the dataset's challenging nature. Key contributions include the atomic action vocabulary, exhaustive labeling protocol with a two-stage annotation process, and analysis of temporal structure via NPMI. The work highlights significant headroom for advances in motion understanding, interaction modeling, and long-range temporal reasoning to achieve robust real-world action recognition and localization.

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 430 15-minute video clips, where actions are localized in space and time, resulting in 1.58M 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. We will release the dataset publicly. 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 15.6% mAP, underscoring the need for developing new approaches for video understanding.

Figures (14)

• Figure 1: The bounding box and action annotations in sample frames of the AVA dataset. Each bounding box is associated with 1 pose action (in orange), 0--3 interactions with objects (in red), and 0--3 interactions with other people (in blue). Note that some of these actions require temporal context to accurately label.
• Figure 2: This figure illustrates the hierarchical nature of an activity. From Barker and Wright Barker1954, pg. 247.
• Figure 3: User interface for action annotation. Details in Sec \ref{['sec:action_label']}.
• Figure 4: We show examples of how atomic actions change over time in AVA. The text shows pairs of atomic actions for the people in red bounding boxes. Temporal information is key for recognizing many of the actions and appearance can substantially vary within an action category, such as opening a door or bottle.
• Figure 5: Sizes of each action class in the AVA train/val dataset sorted by descending order, with colors indicating action types.