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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

We present Recurrent Vision Transformers (RVTs), a novel backbone for objectdetection with event cameras. Event cameras provide visual information withsub-millisecond latency at a high-dynamic range and with strong robustnessagainst motion blur. These unique properties offer great potential forlow-latency object detection and tracking in time-critical scenarios. Priorwork in event-based vision has achieved outstanding detection performance butat the cost of substantial inference time, typically beyond 40 milliseconds. Byrevisiting the high-level design of recurrent vision backbones, we reduceinference time by a factor of 6 while retaining similar performance. To achievethis, we explore a multi-stage design that utilizes three key concepts in eachstage: First, a convolutional prior that can be regarded as a conditionalpositional embedding. Second, local and dilated global self-attention forspatial feature interaction. Third, recurrent temporal feature aggregation tominimize latency while retaining temporal information. RVTs can be trained fromscratch to reach state-of-the-art performance on event-based object detection -achieving an mAP of 47.2% on the Gen1 automotive dataset. At the same time,RVTs offer fast inference (<12 ms on a T4 GPU) and favorable parameterefficiency (5 times fewer than prior art). Our study brings new insights intoeffective design choices that can be fruitful for research beyond event-basedvision.

Recurrent Vision Transformers for Object Detection with Event Cameras