Title:
THREE-DIMENSIONAL AUTO-ENCODER AND TRAINING METHOD THEREFOR, ELECTRONIC DEVICE, AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2024/016464
Kind Code:
A1
Abstract:
Disclosed are a three-dimensional auto-encoder and a training method therefor, a training method for a three-dimensional vision module, an electronic device, and a storage medium. The three-dimensional auto-encoder comprises an encoder and a decoder; the encoder is used for extracting spatial feature parameters of an input picture to be processed and outputting the spatial feature parameters to the decoder; the decoder is used for outputting, according to the spatial feature parameters, a target picture comprising depth information of the picture to be processed; and the picture to be processed and the target picture are used for determining a loss function of the three-dimensional auto-encoder. According to the present invention, perception of three-dimensional scenario features of an image is implemented by means of self-supervision. Moreover, training can be performed by using massive training sets and using a large amount of unlabeled original data, which saves the labor and cost, and a trained model can be conveniently transplanted to other scenarios. Inductive bias of a three-dimensional spatial relationship is added to a neural network design, so that interactive information with an environment comprised in an action can be perceived and the accuracy of recognition can be improved.
More Like This:
Inventors:
HAO LIN (CN)
WANG GUOQUAN (CN)
YE DEJIAN (CN)
WANG GUOQUAN (CN)
YE DEJIAN (CN)
Application Number:
PCT/CN2022/120231
Publication Date:
January 25, 2024
Filing Date:
September 21, 2022
Export Citation:
Assignee:
CLEARTV CO LTD (CN)
ZHEJIANG CLEARTV CO LTD (CN)
ZHEJIANG CLEARTV CO LTD (CN)
International Classes:
G06T7/50; G06N3/04
Foreign References:
| CN114418069A | 2022-04-29 | |||
| CN113628216A | 2021-11-09 |
Other References:
KAUST IBRAHEEM ALHASHIM, WONKA PETER: "High Quality Monocular Depth Estimation via Transfer Learning", ARXIV:1812.11941V2, 10 March 2019 (2019-03-10), XP093130246, Retrieved from the Internet [retrieved on 20240212]
CHAN ERIC R.; LIN CONNOR Z.; CHAN MATTHEW A.; NAGANO KOKI; PAN BOXIAO; DE MELLO SHALINI; GALLO ORAZIO; GUIBAS LEONIDAS; TREMBLAY J: "Efficient Geometry-aware 3D Generative Adversarial Networks", 2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), IEEE, 18 June 2022 (2022-06-18), pages 16102 - 16112, XP034195621, DOI: 10.1109/CVPR52688.2022.01565
CHAN ERIC R.; LIN CONNOR Z.; CHAN MATTHEW A.; NAGANO KOKI; PAN BOXIAO; DE MELLO SHALINI; GALLO ORAZIO; GUIBAS LEONIDAS; TREMBLAY J: "Efficient Geometry-aware 3D Generative Adversarial Networks", 2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), IEEE, 18 June 2022 (2022-06-18), pages 16102 - 16112, XP034195621, DOI: 10.1109/CVPR52688.2022.01565
Attorney, Agent or Firm:
SHANGHAI BESHINING LAW OFFICE (CN)
Download PDF: