Title:
IMAGE PROCESSING METHOD BASED ON PRE-TRAINED LARGE MODEL, AND RELATED APPARATUS
Document Type and Number:
WIPO Patent Application WO/2023/142918
Kind Code:
A1
Abstract:
Disclosed in the present application are an image processing method based on a pre-trained large model, and a related apparatus. The method comprises: acquiring a feature image of a training image by means of a generative network, wherein the resolution of the feature image is the same as the resolution of the training image; fusing the training image with the feature image to obtain a target image; inputting the target image into a pre-trained large model to obtain a processing result; and insofar as parameters of the pre-trained large model remain unchanged, updating parameters of the generative network according to the processing result. In the present application, a generative network is configured for each downstream task, and during a training process, insofar as parameters of a pre-trained large model remain unchanged, parameters of the generative network are updated, thereby reducing the training overheads when the pre-trained large model is deployed on the downstream task, and reducing the model training difficulty and the iterative update difficulty.
Inventors:
CHANG JIANLONG (CN)
ZHANG HENGHENG (CN)
CHEN XIN (CN)
SHI JIAXIN (CN)
WANG ZHIYU (CN)
NING KE (CN)
TIAN QI (CN)
ZHANG HENGHENG (CN)
CHEN XIN (CN)
SHI JIAXIN (CN)
WANG ZHIYU (CN)
NING KE (CN)
TIAN QI (CN)
Application Number:
PCT/CN2023/070316
Publication Date:
August 03, 2023
Filing Date:
January 04, 2023
Export Citation:
Assignee:
HUAWEI CLOUD COMPUTING TECH CO LTD (CN)
International Classes:
G06V10/764
Foreign References:
| CN114648650A | 2022-06-21 | |||
| CN114120032A | 2022-03-01 | |||
| CN113065635A | 2021-07-02 | |||
| CN113947196A | 2022-01-18 | |||
| CN113486162A | 2021-10-08 | |||
| US20210319266A1 | 2021-10-14 |
Attorney, Agent or Firm:
SHENPAT INTELLECTUAL PROPERTY AGENCY (CN)
Download PDF: