Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
IMAGE SEGMENTATION MODEL TRAINING METHOD AND APPARATUS, AND IMAGE SEGMENTATION METHOD AND APPARATUS
Document Type and Number:
WIPO Patent Application WO/2023/230936
Kind Code:
A1
Abstract:
An image segmentation model training method and apparatus, and an image segmentation method and apparatus. The method comprises: acquiring a training sample, the training sample comprising a training image, the training image comprising a labeled mask and a labeled trimap, and the training image being obtained by stitching an image to be processed and a trimap corresponding to the image to be processed; inputting the training sample into an image segmentation model to obtain a predicted trimap corresponding to the image to be processed and a target predicted mask; determining a target loss of the image segmentation model according to the predicted trimap and the target predicted mask; and training the image segmentation model according to the target loss. The trimap corresponding to the training sample does not need a manually generated high-precision trimap, thereby improving the robustness of the trimap required by the training sample, and the optimization target of the image segmentation model is determined according to the predicted trimap and the target predicted mask in a model training process, thereby improving the training effect and robustness of the image segmentation model.

Inventors:
SHI AIJUN (CN)
Application Number:
PCT/CN2022/096495
Publication Date:
December 07, 2023
Filing Date:
May 31, 2022
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
BEIJING XIAOMI MOBILE SOFTWARE CO LTD (CN)
BEIJING XIAOMI PINECONE ELECTRONICS CO LTD (CN)
International Classes:
G06T7/11; G06N3/08
Foreign References:
CN112990331A2021-06-18
CN112541927A2021-03-23
US10304193B12019-05-28
CN112766244A2021-05-07
CN114445376A2022-05-06
CN114332563A2022-04-12
Attorney, Agent or Firm:
FASHENG INTELLECTUAL PROPERTY COMPANY, LTD. (CN)
Download PDF: