用于算网融合环境中面向实时视频流应用的推断加速方法
By modifying deep neural networks to be deployed in a multi-scale block parallel manner in a computing-network convergence environment, the problem of insufficient computing resources was solved, and high-quality, low-latency real-time video streaming transmission was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2023-02-20
- Publication Date
- 2026-07-17
AI Technical Summary
In a computing-network converged environment, existing super-resolution models suffer from high computational load and limited modes, resulting in prolonged end-to-end latency, high consumption of computing resources, low response speed, and an inability to effectively utilize edge computing capabilities.
The deep neural network model is modified by adopting the idea of multi-scale feature extraction, which divides it into multiple independent blocks and deploys them in a distributed manner in a computing network convergence environment. Combined with latency modeling and inference to accelerate decision-making and optimize the utilization of computing resources.
Significantly reduces end-to-end latency, improves resource utilization, adapts to the real-time video streaming application needs in different scenarios, and achieves high-quality, low-latency video transmission.
Smart Images
Figure CN116339977B_ABST