English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Pose-based action recognition is predominantly tackled by approaches whichtreat the input skeleton in a monolithic fashion, i.e. joints in the pose treeare processed as a whole. However, such approaches ignore the fact that actioncategories are often characterized by localized action dynamics involving onlysmall subsets of part joint groups involving hands (e.g. `Thumbs up') or legs(e.g. `Kicking'). Although part-grouping based approaches exist, each partgroup is not considered within the global pose frame, causing such methods tofall short. Further, conventional approaches employ independent modalitystreams (e.g. joint, bone, joint velocity, bone velocity) and train theirnetwork multiple times on these streams, which massively increases the numberof training parameters. To address these issues, we introduce PSUMNet, a novelapproach for scalable and efficient pose-based action recognition. At therepresentation level, we propose a global frame based part stream approach asopposed to conventional modality based streams. Within each part stream, theassociated data from multiple modalities is unified and consumed by theprocessing pipeline. Experimentally, PSUMNet achieves state of the artperformance on the widely used NTURGB+D 60/120 dataset and dense joint skeletondataset NTU 60-X/120-X. PSUMNet is highly efficient and outperforms competingmethods which use 100%-400% more parameters. PSUMNet also generalizes to theSHREC hand gesture dataset with competitive performance. Overall, PSUMNet'sscalability, performance and efficiency makes it an attractive choice foraction recognition and for deployment on compute-restricted embedded and edgedevices. Code and pretrained models can be accessed athttps://github.com/skelemoa/psumnet

PSUMNet: Unified Modality Part Streams are All You Need for Efficient Pose-based Action Recognition