Characteristic extraction method, system and device based on sub-channel quantization and storage medium

Abstract

The embodiment of the invention relates to the technical field of computer vision, and discloses a feature extraction method, system and device based on sub-channel quantization and a storage medium.The method comprises the steps: firstly obtaining a to-be-processed image; inputting the to-be-processed image into the target feature extraction model to obtain target object features output by the target feature extraction model, wherein the target feature extraction model is a compressed feature extraction model obtained by performing model compression on floating point model parameters in theinitial feature extraction model in a mode of determining quantized decimal digits through different channels. Therefore, according to the embodiment of the invention, the initial feature extraction model is subjected to model compression operation, so the compressed model can adapt to the low-power-consumption equipment, and the technical problem that the low-power-consumption equipment is difficult to adapt to the floating point model is solved. Meanwhile, the quantized decimal digits are processed in a sub-channel mode, quantization losses can be reduced, and therefore the current situationthat the network performance is reduced can be relieved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a feature extraction method, system, device and storage medium based on sub-channel quantization. Background technique [0002] With the gradual development of computer vision, especially the continuous development of computer vision based on convolutional neural network, its application scenarios are becoming more and more extensive. [0003] However, considering that deep learning methods based on convolutional neural networks generally require deep network structures and huge model parameters, certain requirements are also placed on the amount of calculation. [0004] In order to cope with this computing demand, although it is possible to perform calculations on a graphics processing unit (GPU, Graphics Processing Unit) and a high-performance central processing unit (CPU, central processing unit), for most embedded devices, it is difficult to directly Deploy neural ne...

AI Technical Summary

Problems solved by technology

[0008] In order to solve the technical problem of difficult adaptation between low-power hardware devices and floating-point models, embo

Images