Data compression method, device and apparatus and storage medium

Abstract

The invention discloses a data compression method, device and apparatus and a storage medium, and the method comprises the steps: compressing data from a high-semantic-redundancy domain to a low-semantic-redundancy domain, obtaining compressed data expressed in a natural semantic form, carrying out lossy coding or lossless coding of the compressed data, obtaining the coded compressed data, and carrying out lossy coding or lossless coding of the compressed data; decoding the encoded compressed data to obtain low-semantic redundant data which can be expressed in a natural semantic form, and decompressing the low-semantic redundant data which can be expressed in the natural semantic form to obtain high-semantic redundant data. According to the data compression method disclosed by the invention, the data with high semantic redundancy is compressed into the data with low semantic redundancy, so that not only is the compression of the data realized, but also the fidelity of the semantic level of the data is emphatically optimized, and the obtained compressed data is the data which can be understood by people and is expressed in a natural semantic form, and can be oriented to one or moremachine analysis tasks. The invention can also be applied to a semantic monitoring scene.

Description

technical field [0001] The present invention relates to the technical field of data processing, in particular to a data compression method, device, equipment and storage medium. Background technique [0002] The traditional data compression framework mainly focuses on the fidelity of the signal level and the subjective fidelity of human eyes. However, in recent years, the application scenarios of machine analysis that have gradually increased mainly require the fidelity of the semantic level of data. Therefore, with the recent Years of machine analysis tasks are gradually applied to production and life, and traditional compression frameworks cannot achieve optimal performance in the application scenarios of machine analysis. [0003] The compression framework for machine analysis proposed in the prior art includes a final layer feature compression framework and an intermediate layer feature compression framework. The final layer feature compression framework compresses and t...

AI Technical Summary

Problems solved by technology

[0004] The feature compression of the middle layer uses the middle layer features of the network instead of the final layer, and takes advantage of the weak task specificity of the middle layer features, so that the compressed and transmitted feature vectors can get the analysis results of multiple tasks in the computing center. At the same time, it can By adjusting the position of the middle layer where the eigenvector is located, the load of the analysis task on the front end and the computing center can be reasonably distributed. The disadvantage of this compression method is that the eigenvector is not understandable by humans.

In the semantic monitoring scenario, that is, we need to monitor some semantic information of the data in real time instead of the original data, this middle layer feature compression method will not meet the demand

Images