Method and device for assessing neural network performance

Abstract

The invention discloses a method and a device for assessing neural network performance. The method comprises the steps of obtaining an evaluation parameter set, a training image set and a to-be-evaluated neural network set; wherein the evaluation parameter set comprises a target sampling rate and a target period number; wherein the evaluation parameter set is obtained through a preset rule; wherein the preset rule is used for guiding a user to adjust a reference parameter set to obtain the evaluation parameter set; using the target sampling rate to sample from the training image set to obtaina sample image set, using the sample image set to train the to-be-evaluated neural network set for N cycles to obtain a trained to-be-evaluated neural network set, and obtaining an evaluation result based on the trained to-be-evaluated neural network set; wherein N is the target period number. The invention further discloses a corresponding device. According to the preset rule provided by the embodiment, the parameters in the reference parameter set are adjusted, and the evaluation accuracy can be improved while the evaluation time is shortened.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a method and device for evaluating the performance of a neural network. Background technique [0002] Neural networks are widely used in the field of computer vision. By training neural networks, neural networks can complete tasks such as image classification and image recognition. However, neural networks with different structures have different performances, and the accuracy of performing the above tasks is also limited. different. Therefore, it is necessary to determine a neural network structure with good performance for the task to be performed. [0003] By evaluating the performance of the neural networks in the neural network library to be evaluated, the performance rankings (ie evaluation results) of the neural networks with different structures in the neural network library to be evaluated can be determined, and then the neural network with better per...

AI Technical Summary

Problems solved by technology

However, the more abundant the training data, the greater the number of training cycles, which means that the required computing resources are large and the training time is very long. Therefore, how to reduce the computing resources consumed to evaluate the performance of the neural network and shorten the cost of evaluating the neural network The training time becomes very meaningful

Images