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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In video transformers, the time dimension is often treated in the same way asthe two spatial dimensions. However, in a scene where objects or the camera maymove, a physical point imaged at one location in frame $t$ may be entirelyunrelated to what is found at that location in frame $t+k$. These temporalcorrespondences should be modeled to facilitate learning about dynamic scenes.To this end, we propose a new drop-in block for video transformers --trajectory attention -- that aggregates information along implicitly determinedmotion paths. We additionally propose a new method to address the quadraticdependence of computation and memory on the input size, which is particularlyimportant for high resolution or long videos. While these ideas are useful in arange of settings, we apply them to the specific task of video actionrecognition with a transformer model and obtain state-of-the-art results on theKinetics, Something--Something V2, and Epic-Kitchens datasets. Code and modelsare available at: https://github.com/facebookresearch/Motionformer

Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers