Dayu v1.1 Released

Dayu system Version 1.1 is released!

Breaking Changes

The basic structure of tasks in dayu is updated from linear pipeline to topological dag (directed acyclic graph) to support more complicated application scenarios.

Features

• A brand-new forwarding mechanism in the dayu system for tasks with dag structure, including splitting nodes with forking and merging nodes with redis.
• A fine-grained and flexible deployment and offloading mechanism for topological logic nodes and physical nodes, which separates the process of model deployment and task offloading and allows collaboration among multi-edges and cloud.
• A more flexible visualization module in frontend to display customized visualization views for system analysis.
• Add our work on model evolution, adaptively switch models based on scenarios. (link)
• Add our latest work on video encoding: CRAVE (Collaborative Region-aware Adaptive Video Encoding). It is a region-adaptive video encoding algorithm for cloud-edge collaborative object detection. (link)

Bug Fix

• Fix problem of write queue full in rtsp datasource server (datasource).
• Fix possible task loss in the system (controller / distributor).
• Add optional cloud/edge parameters filling in template files for flexible parameter specification in cloud-edge pods.

Minor Update

• Add cloud and edge template supplementary to support heterogeneous parameters (backend).
• Beatify frontend pages (frontend).
• Refactor template directory to simplify file structure.
• Unify the base image for system components.
• Add application of age classification. (Current available applications: car-detection, face-detection, gender-classification, age-classification)

For more details, please refer to dayu-v1.1.