HAA500: Human-Centric Atomic Action Dataset with Curated Videos

Abstract

We contribute HAA500 1 , a manually annotated humancentric atomic action dataset for action recognition on 500 classes with over 591k labeled frames. Unlike existing atomic action datasets, where coarse-grained atomic actions were labeled with action-verbs, e.g., "Throw", HAA500 contains fine-grained atomic actions where only consistent actions fall under the same label, e.g., "Baseball Pitching" vs "Free Throw in Basketball", to minimize ambiguities in action classification. HAA500 has been carefully curated to capture the movement of human figures with less spatio-temporal label noises to greatly enhance the training of deep neural networks. The advantages of HAA500 include: 1) human-centric actions with a high average of 69.7% detectable joints for the relevant human poses; 2) each video captures the essential elements of an atomic action without irrelevant frames; 3) fine-grained atomic action classes. Our extensive experiments validate the benefits of human-centric and atomic characteristics of HAA, which enables the trained model to improve prediction by attending to atomic human poses. We detail the HAA500 dataset statistics and collection methodology, and compare quantitatively with existing action recognition datasets.

Few-shot Fine-Grained Action Recognition via Bidirectional Attention and Contrastive Meta-Learning

“…Fine-grained action recognition. Fine-grained action recognition aims to distinguish between sub-classes below basic human action taxonomy [6,33,36]. Compared with coarse-grained actions, the inter-class variance of fine-grained actions is relatively low, which requires differentiating very subtle motion details.…”

“…Compared with coarse-grained actions, the inter-class variance of fine-grained actions is relatively low, which requires differentiating very subtle motion details. Efforts have been made towards building ideal datasets for fine-grained action recognition over the years [6,24,31,33,36,46]. Fine-grained actions in individual sports like taichi [36], baseball [24] are collected at first.…”

“…Fine-grained actions in individual sports like taichi [36], baseball [24] are collected at first. Recently, large-scale datasets with a great number of action classes like FineGym [33] and HAA500 [6] are proposed.…”

“…We evaluate our model and the competitor methods on two recently large-scale fine-grained action recognition datasets, namely FineGym [33] and HAA500 [6]. All models are tested on 5-way 1/3/5-shot recognition tasks.…”

“…Since training and test protocols are quite influential on the result of few-shot recognition problems [37,48], we establish the first benchmark for few-shot fine-grained action recognition on two recently proposed large-scale fine-grained action datasets, i.e. FineGym [33] and HAA500 [6]. We evaluate our method and various challenging competitors under specific protocols and make them available to facilitate future works.…”