Image recognition model training method and device, image recognition method and device and equipment

Abstract

The invention relates to an image recognition model training method and device, an image recognition method and device and equipment. In the training process of the image recognition model, a weight coefficient is adaptively and dynamically adjusted through a Frank-Wolfe algorithm; and the training of the image recognition model is realized through the alternate execution of the model parameter updating stage and the weight coefficient updating stage, so that the automation degree of the training of the image recognition model can be improved and the adjustment cost of the weight coefficient can be reduced. The weight coefficient is dynamically adjusted through the Frank-Wolfe algorithm; the influence of uncertainty of manual adjustment of the weight coefficient can be avoided, it can be guaranteed that the weight coefficient is continuously optimized, a second image recognition model can reach the image recognition accuracy similar to that of a first image recognition model, and the image recognition accuracy in the image processing technology can be improved in practical application.

Description

technical field [0001] The present disclosure relates to the technical field of artificial intelligence, and in particular to an image recognition model training, image recognition method, device and equipment. Background technique [0002] At present, more and more fields need to use image recognition models for image recognition processing. In order to meet this demand, knowledge distillation technology is often used to compress the trained high-complexity image recognition models (teacher models) to obtain student In this way, an image recognition model with small parameters, simple structure, and easy deployment can be obtained efficiently, so that it can efficiently meet the needs of image recognition models in various fields. However, most of the current knowledge distillation technologies manually adjust the weight coefficients of the loss function, which will increase the cost and difficulty of student model training, and the manual adjustment method will lead to low...

AI Technical Summary

Problems solved by technology

However, most of the current knowledge distillation techniques manually adjust the weight coefficients of the loss function, which will increase the cost and difficulty of student model training, and the manual adjustment method will lead to low image recognition accuracy of the student model.

Images