Title:
PROCESSING METHOD AND APPARATUS BASED ON FACIAL RECOGNITION, AND DEVICE AND READABLE STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2020/093303
Kind Code:
A1
Abstract:
A processing method and apparatus based on facial recognition, and a device and a readable storage medium. The method comprises: first using a facial data set and a Softmax-based first loss function to train a facial feature extraction model (S101), and then using a person identity data set and a Triplet-based second loss function to carry out model re-training on the facial feature extraction model to obtain a final facial feature extraction model (S102). The convergence speed when a model is trained using a person identity data set and a Triplet-based second loss function can be increased, and model training efficiency can be improved; and even if there are fewer person identity data sets, a facial feature model can also be better trained using these person identity data sets by means of the Triplet-based second loss function, and data characteristics of a facial data set and the person identity data sets can be fully used, thereby improving the precision of extracting facial features by means of the facial feature model when being applied to a "person identity scenario" and being able to improve the accuracy of facial recognition.
Inventors:
WU XIAOMIN (CN)
Application Number:
PCT/CN2018/114537
Publication Date:
May 14, 2020
Filing Date:
November 08, 2018
Export Citation:
Assignee:
BITMAIN TECH INC (CN)
International Classes:
G06K9/00
Foreign References:
| CN107871100A | 2018-04-03 | |||
| CN107103281A | 2017-08-29 | |||
| CN106203533A | 2016-12-07 | |||
| CN108197561A | 2018-06-22 |
Other References:
SONG, YILONG ET AL.: "Deep Learning Facial Feature Extraction Method Based on Mixed Training", NEW TECHNOLOGY & NEW PROCESS, 25 March 2018 (2018-03-25), pages 39 - 42, ISSN: 1003-5311
Attorney, Agent or Firm:
LEADER PATENT & TRADEMARK FIRM (CN)
Download PDF: