Image classification network compression method based on parameter reinitialization
A technology for re-initializing and classifying networks, applied in neural learning methods, biological neural network models, instruments, etc., can solve problems such as wrong pruning, overall model performance degradation, unfavorable pruning schemes, etc., to improve performance and improve performance. Effect
Pending Publication Date: 2022-02-15
NORTHWESTERN POLYTECHNICAL UNIV
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
This method relies on sparse training to identify important convolutional channels, and unimportant channels will always be suppressed during sparse training, and it is difficult to play an important role under the action of sparse regularization, which is easy to cause wrong pruning. It is beneficial to find the optimal pruning scheme, which reduces the overall performance of the model
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0029] In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
[0030] For the convolutional layer in the neural network, the calculation process is:
[0031] Y=X*w (1)
[0032] In the formula, is the input feature map tensor, is the output feature map tensor, is the convolution weight parameter, c and n are the number of input and output channels respectively, h and w are the height and width of the input feature map respectively, h′ and w′ are the height and wid...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention relates to an image classification network compression method based on parameter reinitialization, and belongs to the technical field of image processing and recognition. According to the method, in a network pre-training process, parameter reinitialization is performed on unimportant channels once every t times of iteration on a complete training data set; and in a network training process, the parameter reinitialization is performed on the unimportant channels, so that more filter forms can be introduced for a model, convolution channels which are wrongly pruned can be reactivated, and performance of the compressed network model can be improved. When the method is applied to an image classification task, the number of parameters and the amount of operation of a model can be reduced on the premise of ensuring the classification accuracy, and the method can be conveniently used in mobile equipment such as a mobile phone.
Description
technical field [0001] The invention belongs to the technical field of image processing and recognition, in particular to an image classification network compression method based on parameter reinitialization. Background technique [0002] With the development of neural network technology, more and more network models have replaced traditional artificial models, and have achieved great success in machine vision fields such as image classification and face recognition. The convolutional neural network model relies on a huge training data set and a large number of complex operations to extract image features. After repeated iterative training, the model's high abstraction ability for the essential features of the target is improved, and a robust recognition effect is achieved. [0003] However, due to the huge size of the neural network model with superior performance, it also occupies a large amount of storage space and computing resources. Terminal devices with limited reso...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More IPC IPC(8): G06V10/764G06V10/82G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/241
Inventor 苏雨刘广哲张科王靖宇
Owner NORTHWESTERN POLYTECHNICAL UNIV