Title:
IMAGE PROCESSING, NETWORK TRAINING AND ENCODING METHODS, APPARATUS, DEVICE, STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2022/217725
Kind Code:
A1
Abstract:
Image processing, network training and encoding methods, an apparatus, a device, and a storage medium. The network training method comprises: performing joint training on a codec network and a task execution network according to a joint loss function of the codec network and the task execution network and a preset sample training set, until the value of the joint loss function meets a convergence condition, so as to obtain a trained codec network and a trained task execution network. The task execution network is used for performing a machine learning task according to a decoded reconstructed image outputted by the codec network, and the joint loss function comprises a loss function of the task execution network and a function representing the bit rate of a feature stream of an input image of the codec network.
Inventors:
YU LU (CN)
ZHU BINGJIE (CN)
DAI ZHENYU (CN)
ZHU BINGJIE (CN)
DAI ZHENYU (CN)
Application Number:
PCT/CN2021/098413
Publication Date:
October 20, 2022
Filing Date:
June 04, 2021
Export Citation:
Assignee:
UNIV ZHEJIANG (CN)
GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD (CN)
GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD (CN)
International Classes:
G06K9/62
Foreign References:
| CN111797855A | 2020-10-20 | |||
| CN111968044A | 2020-11-20 | |||
| US20190130212A1 | 2019-05-02 | |||
| US5986594A | 1999-11-16 |
Other References:
CHENG ZSUN HTAKEUCHI M ET AL.: "Learned image compression with discretized gaussian mixture likelihoods and attention modules[C", PROCEEDINGS OF THE IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, 2020, pages 7939 - 7948
YUXIN WUALEXANDER KIRILLOVFRANCISCO MASSAWAN-YEN LOROSS GIRSHICK, DETECTRON2. [OL, 2019, Retrieved from the Internet
YUXIN WUALEXANDER KIRILLOVFRANCISCO MASSAWAN-YEN LOROSS GIRSHICK, DETECTRON2. [OL, 2019, Retrieved from the Internet
Attorney, Agent or Firm:
CHINA PAT INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF: