A Research on Face Recognition Open Source Development Framework Based on PyTorch

Autor: Xiaofu Du, Wenkai Zang, Hedan Liu, Xinghan Huang
Rok vydání: 2021
Předmět:
Zdroj: 2021 International Symposium on Computer Technology and Information Science (ISCTIS).
Popis: At present, there are some problems in the development framework of various neural networks for face recognition in the laboratory environment, for example, the development environment is not standard, the dataset is not standard, and it is not friendly to beginners. In order to solve the above problems, this paper presents an open-source face recognition research and development framework based on PyTorch to help beginners quickly complete the experimental environment. First, based on the current mainstream CUDA technology, combined with PyTorch machine learning library, and with general third-party libraries such as OpenCV, a face recognition research and development framework is built. Then, a custom image data collection acquisition mechanism with three-tier filtering capabilities is proposed to help users quickly obtain high-quality datasets for use by the R&D framework. In addition, based on this research and development framework, some methods such as model compression, parameter sharing, were used to speed up model training. Finally, two classical experiments of handwritten numeric recognition and face recognition are implemented in this research and development framework. The test results show that the R&D framework is easy to use and has high practical value.
Databáze: OpenAIRE