Model compression method and device, computing equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a model compression method and device, equipment and a storage medium. The model compression method comprises the following steps: acquiring a test image and a to-be-compressed model, wherein the to-be-compressed model comprises a plurality of cascaded feature extraction layers; inputting a test image into each feature extraction layer, and performing feature extraction on the test image through a filter to obtain a multi-channel feature map, wherein each feature channel in the multi-channel feature map corresponds to one sub-feature map; converting the sub-feature map into a visual feature map, and determining the importance of a feature channel corresponding to the sub-feature map based on a plurality of feature values in the visual feature map; determining a target pruning channel according to the importance degree of the feature channel; and performing pruning processing on the target pruning channel to obtain a compressed target model. According to the method, the model pruning accuracy can be effectively improved, accurate compression of the model is realized, the model calculation amount can be effectively reduced, and deployment is convenient.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a model compression method, device, computing equipment and storage medium. Background technique [0002] In recent years, models based on convolutional neural networks have performed well in many tasks, but these models require a large amount of computational overhead, and these models often contain a large amount of redundant information, so model compression becomes essential. step. Commonly used model compression methods include model pruning, quantization, and distillation. [0003] For the current model pruning operation, the main problem is how to select relatively unimportant feature channels and remove them. The existing methods for how to select relatively unimportant convolution kernels are generally as follows: one is to use the proportion of zero values ​​​​in the feature map, and then use the feature channel corresponding to the feature map...

AI Technical Summary

Problems solved by technology

[0004] Embodiments of the present invention provide a model compression method, device, computing device, and storage medium to solve the current problem of low accuracy of model pruning due to the importance of evaluating feature channels based on numerical values ​​in feature maps

Images