Pedestrian generation method based on pedestrian mask and multi-scale discrimination

A multi-scale, pedestrian technology, applied in neural learning methods, character and pattern recognition, biological neural network models, etc., can solve the problems of blurred pedestrian outline and background boundary, unclear pedestrian image, etc., and achieve clear pedestrian outline and background boundary. Effect

Active Publication Date: 2020-09-04
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF9 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The pedestrian image output by the conventional image processing method is relatively unclear, and the boundary between the outline of the pedestrian and the background is relatively blurred

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
  • Pedestrian generation method based on pedestrian mask and multi-scale discrimination
  • Pedestrian generation method based on pedestrian mask and multi-scale discrimination
  • Pedestrian generation method based on pedestrian mask and multi-scale discrimination

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0043] This embodiment provides a pedestrian generation method based on pedestrian mask and multi-scale discrimination, such as figure 1 As shown, a network structure including a generator and three discriminators is constructed in this method. Feed the pedestrian mask and noise into the generator to generate pedestrian images. The discriminator receives the pedestrian mask and the pedestrian image, judges the authenticity of the image pair, and feeds the discriminative information to the generator through the loss function to guide the generator to generate pedestrians. Through the continuous game between the generator and the discriminator, finally generate Model pose features of pedestrians.

[0044]The generator uses u-Net structure, including encoder and decoder. Such as image 3 As shown, the encoder performs downsampling in the form of convolution to encode the input image; the decoder performs upsampling and restoration in the form of deconvolution to generate an im...

Embodiment 2

[0050] Further improvement based on Example 1, the loss function includes generative confrontation loss, L1 loss function, feature matching loss and total loss:

[0051] (1) The generative adversarial loss in this method guides the generation of images. Among them, G represents the generator, D represents the discriminator, s is the pedestrian mask, x is the real pedestrian, and z is the noise vector. G(s,z) is the generated image, D k Denotes the kth discriminator.

[0052] L GAN (G,D)=E s,x [logD(s,x)]+E s,z [log(1-D(s,G(s,z)))];

[0053]

[0054] (2) This method uses the L1 loss function to ensure the consistent mapping relationship between the input image and the generated image.

[0055] L L1 (G)=E s,x,z [||y-G(s,z)|| 1 ];

[0056] L GAN (G,D)=E (s,x) [logD(s,x)]+E s [log(1-D(s,G(s))];

[0057] (3) This method uses feature matching loss L FM (G,D k ) to improve the stability of network training; the main goal of this loss function is to extract features...

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 invention discloses a pedestrian generation method based on a pedestrian mask and multi-scale discrimination. The pedestrian generation method specifically comprises the following steps that the pedestrian mask is input into a generator to generate a pedestrian image; the discriminator discriminates an image pair formed by the pedestrian mask and the generated pedestrian image; and the discriminator returns a discrimination result to the generator to guide the generator to continue to generate a pedestrian image more conforming to the posture of the pedestrian mask. According to the method, a multi-scale discrimination mode is used, a plurality of discriminators are included, and pedestrian masks of different scales and image pairs composed of generated pedestrians are discriminated respectively. The method proves that the method based on the pedestrian mask and multi-scale discrimination can generate the pedestrian with a specific mask attitude, the generated image is finer, and the pedestrian contour and the background boundary are clearer.

Description

technical field [0001] The invention relates to the technical field of image generation, in particular to a method for generating pedestrians based on pedestrian masks and multi-scale discrimination. Background technique [0002] Many problems in image processing are to convert an input image into a corresponding output image, such as the conversion between grayscale images, gradient images, and color images. Usually each problem uses a specific algorithm. For example, when using CNN to solve image conversion problems, a specific loss function should be set according to each problem to allow CNN to optimize. The essence of these methods is actually the mapping from pixel to pixel. "Translation" is often used for translation between languages, such as translation between Chinese and English. But image translation means converting from image to image in different forms. For example, a scene can be converted into an RGB full-color image, a sketch, or a grayscale image. A ni...

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): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V40/103G06N3/045G06F18/214
Inventor 匡平肖小霞
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap