Real-time video quality optimization-oriented deep hybrid model traffic control method and device, and storage medium

Abstract

The invention provides a real-time video quality optimization-oriented deep hybrid model traffic control method and a device, and a storage medium, and the method comprises the steps: obtaining a network information state of a network transmission data packet in real time, obtaining a first feature set from the network information state, inputting the first feature set to a trained first intelligent congestion control model, outputting a probability feature vector from a feature layer of the model, wherein the model is obtained by taking a prediction result of a congestion control algorithm based on a fixed mapping rule as label training; obtaining a second feature set based on the network information state, inputting the second feature set to a second intelligent congestion control model which is trained online and based on reinforcement learning, and outputting a probability feature vector from a feature layer of the model; performing enhancement operation on the probability feature vector output by the first intelligent congestion control model; fusing the enhanced probability feature vector with the probability feature vector output by the second intelligent congestion control model; and obtaining a predicted code rate based on the fused probability feature vector, and performing flow control.

Description

technical field [0001] The invention relates to the technical field of video network transmission, in particular to a deep hybrid model flow control method, device and storage medium oriented to real-time video quality optimization. Background technique [0002] The network transmission protocol (congestion control algorithm) has a research history of more than 30 years, and is used for real-time control of video transmission traffic by predicting (estimating) the bit rate at the next moment. With the rapid development of Internet applications in recent years, traditional transmission protocols can no longer meet the needs of real-time video applications, such as large bandwidth, low latency, and high definition. Most of the current traditional (non-intelligent) video transmission protocols have the problem of "miscellaneous but not precise", and a series of assumptions will be set for the network's pre-conditions. When these assumptions do not match the new network environm...

AI Technical Summary

Problems solved by technology

However, the inventors of the present invention have found that the two-level hybrid strategy of the Orca algorithm is only a "shallow fusion", and this kind of switching between coarse and fine granularity cannot achieve accurate network link estimation, especially for those with high delay requirements. real-time video transmission application

Images