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

METHOD AND SYSTEM OF ADAPTING AN INITIAL MODEL OF A NEURAL NETWORK, Storage part and vehicle

An initial model and neural network technology, applied in the field of image processing, can solve problems such as time-consuming and semantic segmentation data difficulties

Pending Publication Date: 2021-08-31
TOYOTA JIDOSHA KK +1
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0010] However, obtaining semantically segmented data during these changing weather conditions is particularly difficult and time-consuming

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
  • METHOD AND SYSTEM OF ADAPTING AN INITIAL MODEL OF A NEURAL NETWORK, Storage part and vehicle
  • METHOD AND SYSTEM OF ADAPTING AN INITIAL MODEL OF A NEURAL NETWORK, Storage part and vehicle
  • METHOD AND SYSTEM OF ADAPTING AN INITIAL MODEL OF A NEURAL NETWORK, Storage part and vehicle

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0075] Reference will now be made in detail to the exemplary embodiments of the invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts.

[0076] figure 1 A flowchart of a method for adapting an initial model of a neural network according to an embodiment of the present invention is shown.

[0077] The invention has been implemented using Segnet, MobileNeyV2 and DeepLabV3, but other architectures may be used.

[0078] More precisely, the method of the present invention will utilize a source domain image x obtained under high visibility conditions s The initial model trained Adapted to images of the target domain acquired under low visibility conditions (eg, dark, foggy or snowy conditions).

[0079] At step E10, using the source domain image x obtained under high visibility conditions s Train the initial model

[0080] Such as figu...

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

A method of adapting an initial model trained with a marked image of a source domain to an adaptive model includes: copying the initial model into the adaptive model; dividing the adaptation model into an encoder portion and a second portion, wherein the second portion is configured to process features output from the encoder portion; adapting the adaptation model to the target domain using the images (xs) of the source domain and the target domain while fixing the parameters of the second portion and minimizing a function of the following two distances: the distance between the initial model and the feature of the source domain of the output of the encoder of the adaptation model; and measuring a distance of a distribution distance between probabilities of features obtained for the image of the source domain and the image of the target domain.

Description

technical field [0001] The present invention relates to the field of image processing, and more precisely to the improvement of the classification performance of neural networks. Background technique [0002] The invention finds a dedicated application in the field of image classification for autonomous vehicles, but can be applied to process any type of image. [0003] Semantic information provides a valuable source for understanding the scene around autonomous vehicles in order to plan their actions and make decisions. [0004] Semantic segmentation of these scenes allows the recognition of cars, pedestrians, lanes, etc. Therefore, semantic segmentation is a backbone technology for self-driving systems or other automated systems. [0005] Semantic image segmentation typically uses models such as neural networks to perform segmentation. These models need to be trained. [0006] Training a model typically involves feeding known images to the model. For these images, 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): G06N3/08G06K9/62G06V10/764
CPCG06N3/08G06F18/2415G06V20/58G06V10/82G06V10/764G06N3/088G06F18/22G06F18/214G06F18/2163G06N3/045
Inventor 格布里尔·欧斯梅兹奥瑞厄兹古尔·额肯特克里斯蒂安·劳吉尔
Owner TOYOTA JIDOSHA KK