Model compression method and device, electronic equipment and computer storage medium

A model and network model technology, applied in the field of model compression, can solve the problems of teacher network detection performance gap, single-stage application has not been explored, student network performance is not ideal, etc., to achieve the effect of target detection performance improvement
CN109711544AInactive Publication Date: 2019-05-03BEIJING SENSETIME TECH DEV CO LTD
3 Cites 38 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING SENSETIME TECH DEV CO LTD
Publication Date
2019-05-03
Estimated Expiration
Not applicable Β· inactive patent

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention provides a model compression method and device, electronic equipment and a computer storage medium. The method comprises the steps that training sample data are acquired, and the training sample data comprise label sample data; Respectively training the teacher network model and the student network model by using the training sample data to obtain an adaptive distillation loss function and a focus loss function; and performing back propagation on the student network model according to the adaptive distillation loss function and the focus loss function to obtain a trained studentnetwork.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The present application relates to the field of artificial intelligence, in particular to a model compression method, device, electronic equipment and computer storage medium. Background technique

[0002] In recent years, deep learning networks have achieved great success in object detection applications in the field of computer vision. However, since the deep learning network model often contains a large number of model parameters, the amount of calculation is large, the processing speed is slow, and its application is mostly in the cloud, so it still faces huge challenges in the implementation of the terminal.

[0003] In order to reduce the redundancy of the network model, researchers at home and abroad have proposed a distillation learning algorithm. In distillation learning, the training of the student network model is guided by refining or distilling the knowledge of the teacher network with a complex structure to the student network model with ...

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