Skip to main content

Dayu v1.2 Released

周文晖
周文晖
Dayu Project Maintainer
苏浩洋
苏浩洋
Dayu Project Contributor

Dayu system Version 1.2 is released!

Features​

  • Add a cyclic redeployment mechanism of service processors for further flexible task processing.
  • Add new services to construct a logical topology, including pedestrian-detection, vehicle-detection, exposure-identification, category-classification and license-plate-recognition.

Bug Fix​

  • Fix concurrency conflicts for starting up multiple streams in rtsp video and http video (datasource).
  • Fix frame index updating error in http video source (datasource).
  • Fix port occupation bug of rtsp video datasource after shutting down (datasource).
  • Support dynamic variable update in visualization modules (frontend).
  • Support cloud-only deployment of processors (backend).
  • Fix task forwarding bug in inconsistent deployment and offloading decisions (controller).
  • Fix possible database locking in concurrency access of distributor (distributor).
  • Add cache TTL of Kubernetes configurations to avoid additional expense in PortInfo and KubeConfig.

Minor Update​

  • Update more flexible visualization modules to switch different user-defined configurations in multi-stream scenarios (frontend / backend).
  • Clean up frontend code and beatify frontend pages (frontend).
  • Add persistent storage of installation configuration (frontend).
  • Add system visualization to monitor system parameters, including resource usage, scheduling cost and so on (frontend / backend / scheduler).
  • Improving a fine-grained monitoring architecture including monitoring cpu and gpu flops (monitor).
  • Add 'USE_TENSORRT' option in processors to choose whether using tensorrt mode (processor).
  • Add model flops calculation in processors to support model flops monitoring (processor).
  • Update 'ImagePullPolicy' in template files from 'IfNotPresent' to 'Always' to ensure pulling the latest images.
  • Update detailed overhead time statistics in scheduling logs (scheduler).
  • Accelerate backend visualization through caching mechanism (backend).
  • Add ROI ID to detection/tracking applications to support roi-accelerated classification applications (processor).
  • Add ROI Classifier to accelerate roi-level classification applications (processor).
  • Optimize local service processor by sharing temporary directory on local devices (controller/processor).
  • Add adjustable request scheduling interval in generator to avoid frequent scheduling (generator).
  • Add temporary file cleaning mechanism in local controller to avoid disk occupation (controller).
  • Add compatible docker image building for jp4/jp5/jp6 for processor in Nvidia Jetson devices (processor/monitor).
  • Update scenario data structure and processing (processor).

For more details, please refer to dayu-v1.2.