Cross-platform entropy coding method and decoding method for deep learning coding

Abstract

The invention discloses a cross-platform entropy coding method and decoding method for deep learning coding. The cross-platform entropy coding method comprises the steps of determining and calculatinga network module of an entropy coding model based on deep learning coding; replacing the model parameter in the network module with the model parameter of the floating-point number with the precisiongreater than or equal to float64, and replacing the intermediate variable used for calculation with the intermediate variable of the floating-point number with the precision greater than or equal tofloat64; and loading trained model parameters, and initializing an entropy coding model based on deep learning coding. Through the method, cross-platform use can be achieved, model parameters do not need to be redesigned and retrained, the method can be directly applied to deployment of the model parameters, and performance and efficiency cannot be reduced.

Description

technical field [0001] The invention relates to the technical field of image coding, in particular to a cross-platform entropy coding method and a decoding method for deep learning coding. Background technique [0002] In recent years, image compression methods based on deep learning have attracted the attention of many researchers. Researchers have achieved a lot in this research direction. For example, in 2016, Toderici et al. proposed an image compression framework based on a circular convolutional neural network, which achieved a compression performance that surpassed the most commonly used traditional encoder JPEG. This is the first time that a learning-based method has surpassed traditional methods. In 2018, another encoding framework based on convolutional neural networks achieved performance beyond the best traditional encoding method BPG at the time. These results fully demonstrate the great potential of image coding methods based on deep learning. [0003] The f...

AI Technical Summary

Problems solved by technology

This method can effectively solve the problem of different accuracy of the entropy coding model calculated at the codec end, but the disadvantage of this method is that it needs to redesign the network and retrain the network, which is not applicable to the existing trained models

Images