Unlock instant, AI-driven research and patent intelligence for your innovation.

Semantic segmentation model training method and system based on ohm

A technology for semantic segmentation and model training, applied in character and pattern recognition, instruments, computer parts, etc., can solve the problems of poor segmentation model training results and small proportions, and achieve the effect of improving the effect and training speed.

Inactive Publication Date: 2020-05-01
魔视智能科技(上海)有限公司
View PDF7 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the proportion of pedestrians is relatively small compared to the overall pixels of the picture. This imbalance will make the training results of the segmentation model relatively poor.

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
  • Semantic segmentation model training method and system based on ohm
  • Semantic segmentation model training method and system based on ohm
  • Semantic segmentation model training method and system based on ohm

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0040] like figure 1 As shown, an ohem-based semantic segmentation model training method includes the following steps:

[0041] S101. Calculate the loss function of each category:

[0042] a. Input the sample image into the semantic segmentation model, and obtain the confidence p of each pixel in the image belonging to each category.

[0043] In this embodiment, the semantic segmentation model adopts the Deeplabv3+ model.

[0044] Input the sample image with pixel size of 3*H*W into the Deeplabv3+ model, and you will get an output N*H*W, where H and W are the height and width of the image, and N represents the number of image output channels (channels), that is, the model The number of semantic categories in which the image is segmented, specifically, each output channel represents the confidence of whether each pixel belongs to a certain category, and the value of the confidence is distributed between 0-1.

[0045] b. Take all positive sample pixels for the current categor...

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 relates to a semantic segmentation model training method based on ohm. The method comprises the following steps: 1, calculating a loss function of each category; 2, carrying out the weighted average of the loss function of each category, and obtaining a final segmentation loss function; and 3, performing iterative optimization on the semantic segmentation model by using a gradient back propagation strategy. The method aims at solving the problem that the model effect is poor when the proportion of positive and negative samples in a segmentation task is seriously uneven. Accordingto the novel segmentation model training method, the positive and negative sample proportion is balanced to train the model, the weight of the positive sample in the training process is increased, the difficult sample is preferably selected, and the segmentation task effect and the training speed are improved.

Description

technical field [0001] The invention relates to the technical field of semantic segmentation, in particular to an ohem-based semantic segmentation model training method and system thereof. Background technique [0002] With the development of computer vision, more vision-based perception technologies have been applied in the industry. Commonly used target recognition technologies include target detection, semantic segmentation, instance segmentation, etc., to help tasks locate and identify target information to complete follow-up work, such as Autopilot. How to obtain more accurate bounding boxes and how to obtain more accurate segmentation maps are also important research directions in academia and industry, and many improvement techniques have been proposed. Among them, the semantic detection task refers to inputting a picture, classifying the category of each pixel in the picture, and obtaining a segmentation map corresponding to the original image, so as to complete the...

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): G06K9/62G06K9/00G06K9/34
CPCG06V20/588G06V10/26G06F18/241G06F18/214
Inventor 袁施薇李发成张如高虞正华
Owner 魔视智能科技(上海)有限公司