Image processing method and device, electronic equipment, medium and program product

Abstract

The embodiment of the invention discloses an image processing method and device, electronic equipment, a medium and a program product. The image processing method comprises the following steps: acquiring quantization precision sensitivities of a plurality of layers in a multi-layer neural network for image processing; generating an initial population of an evolutionary algorithm based on the quantization precision sensitivity of the layer and a random number generated for the layer, each member in the initial population corresponding to a quantization precision configuration corresponding to the multilayer neural network; evaluating the performance of the members in the initial population by using an evaluation image set, and updating the population based on an evaluation result until the members with the performance meeting a preset condition appear in the population; and determining the quantization precision of multiple layers in the multi-layer neural network according to the members of which the expressions meet the preset conditions.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular to an image processing method, device, electronic equipment, media, and program product. Background technique [0002] In a cloud computing environment such as a public cloud, when image processing is performed, the proportion of reasoning requirements of the neural network model is increasing, and there is a need to optimize the performance of reasoning tasks. When the neural network model is quantized to a 16-bit floating point (16bit Floating Point, FP16) model, the calculation accuracy can be basically not lost. Neural network model quantization can also be quantized to 8-bit integer (8bit Integer, INT8) type operations. The nominal computing power of INT8 is twice that of FP16, but there is an unacceptable loss of precision if all models are directly quantized to INT8. Therefore, when the quantization accuracy of FP16 does not meet the speed requirements,...

AI Technical Summary

Problems solved by technology

When solving the mixed quantitative optimization solution, the computational resources required for traversal selection of the optimal solution are huge, time-consuming, low in efficiency, and difficult to meet the actual needs. Therefore, it is necessary to obtain a method for quickly solving the mixed quantitative optimization solution.

Images