Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Deep learning model training method and device, training equipment and storage medium

A deep learning and model training technology, applied in the information field, can solve the problems of reduced training efficiency and difficulty in ensuring the processing accuracy of deep learning models

Active Publication Date: 2019-06-21
BEIJING SENSETIME TECH DEV CO LTD
View PDF7 Cites 18 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In some application scenarios, if only unsupervised training can be performed based on the model itself, and the unsupervised training method, on the one hand, it is difficult to ensure the processing accuracy of the deep learning model after it goes online; on the other hand, in order to improve the accuracy, developers There may be a tendency to overtrain, which leads to unnecessary training, which makes training less efficient

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 more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Deep learning model training method and device, training equipment and storage medium
  • Deep learning model training method and device, training equipment and storage medium
  • Deep learning model training method and device, training equipment and storage medium

Examples

Experimental program
Comparison scheme
Effect test

example 1

[0330] The depth model provided in this example is a bottom-up deep learning model. This deep learning model is a two-step multi-target detection: In this example, the target is a human body as an example. For details, see Figure 5A with Figure 5B Shown:

[0331] The first step is to predict the first type of feature map and obtain the position of each key point; for example, the first type of feature map, the first feature map and / or the second feature map can be obtained by using a bottom-up key point model ;

[0332] The second step is to cluster the different features of each key point to obtain a complete human pose. In this example, an auxiliary training module can be used, for example, a clustering module based on human body posture guidance, respectively clustering the first features contained in the first feature map based on the first type feature map, and clustering the first features contained in the first feature map based on the first The class feature map c...

example 2

[0353] Such as Figure 6A As shown, this example provides a deep learning model, including:

[0354] Feature extraction layer, including: multiple convolutional sublayers and pooling layers, in Figure 6AThe number of convolutional sublayers in the middle is 5; the pooling layer is a maximum pooling layer, and the maximum pooling layer here is a downsampling layer that retains the maximum value; the channel number of the first convolutional sublayer is 64. The size of the convolution kernel is 7*7, and the convolution step is 2; the number of channels of the second convolution sublayer is 128, the size of the convolution kernel is 3*3, and the convolution step is 1; The number of channels of the three convolution sublayers is 128, the size of the convolution kernel is 7*7, and the convolution step is 1; the number of channels of the fourth convolution sublayer is 128, and the size of the convolution kernel is 3* 3. The convolution step is 1; the number of channels of the fif...

example 3

[0358] Human key point detection and tracking is the basis of video analysis, and has important application prospects in the field of security and motion analysis. The single-frame multi-person posture detection technology based on the first feature (Keypoint Embedding) is a very advanced bottom-up human key point detection technology.

[0359] This method outputs the first feature map while outputting the first feature map of the human body. The dimensionality of the embedded feature map and the first-class features Figure 1 It can also be represented by a series of two-dimensional matrices of output resolution size, where the category of each key point corresponds to a two-dimensional matrix. The first type of feature map and the embedded feature map can correspond one-to-one in the spatial position. In the output embedding feature map, each key point of the same person has similar embedding values, while the embedding values ​​of key points of different people are very d...

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

No PUM Login to View More

Abstract

The embodiment of the invention discloses a deep learning model training method and device, training equipment and a storage medium. The deep learning model training method comprises the following steps: training a deep learning model by using a training image to obtain training characteristics output by the deep learning model; carrying out conversion processing on the training features by usingan auxiliary training module to obtain conversion features; Determining a loss value based on the conversion feature; And determining whether to continue to train the deep learning model based on theloss value.

Description

technical field [0001] The present invention relates to the field of information technology, in particular to a deep learning model training method and device, training equipment and a storage medium. Background technique [0002] In the field of security and motion analysis, it is necessary to detect targets such as portraits in images, and then perform follow-up processing such as target behavior analysis and target tracking based on target detection. [0003] In related technologies, a deep learning model or the like is generally used to detect objects in an image. Generally, deep learning models need to use training images for training before going online. In some application scenarios, if only unsupervised training can be performed based on the model itself, and the unsupervised training method, on the one hand, it is difficult to ensure the processing accuracy of the deep learning model after it goes online; on the other hand, in order to improve the accuracy, develop...

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
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/62G06K9/00
Inventor 金晟刘文韬钱晨
Owner BEIJING SENSETIME TECH DEV CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com