METHOD AND DEVICE FOR DAMAGE SEGMENTATION OF VEHICLE DAMAGE IMAGE

Title:

METHOD AND DEVICE FOR DAMAGE SEGMENTATION OF VEHICLE DAMAGE IMAGE

Document Type and Number:

WIPO Patent Application WO/2020/052352

Kind Code:

A1

Abstract:

Provided is a method for component segmentation and damage segmentation based on migration learning. The method specifically comprises: in a first aspect, a migration method can be learned based on high-frequency component categories with rich detection and segmentation data, then for a low-frequency component category with only detection data and no rich segmentation data, a corresponding segmentation result can be obtained by the detection frame thereof acting on this migration method; in a second aspect, a segmentation migration detection method can be learned based on high-frequency component categories with rich detection and segmentation data, then for a low-frequency damage category with only detection data and no rich segmentation data, a corresponding segmentation result can be obtained by the detection frame thereof acting on this migration method; and in a third aspect, a segmentation result of damage can be obtained through cross-domain migration learning from components to the damage. By means of these methods, it is possible to solve the problem that it is difficult to acquire segmentation data of a long-tailed component category and damage category.

Inventors:

XU JUAN (CN)

Application Number:

PCT/CN2019/097904

Publication Date:

March 19, 2020

Filing Date:

July 26, 2019

Export Citation:

Click for automatic bibliography generation Help

Assignee:

ALIBABA GROUP HOLDING LTD (CN)

International Classes:

G06T7/12

Domestic Patent References:

WO2017176304A1

2017-10-12

Foreign References:

CN108446618A	2018-08-24
CN107392218A	2017-11-24
CN107977895A	2018-05-01

Other References: