Research

The Visual Big Data Research Center studies facial recognition algorithms, algorithms for discriminating various facial attributes such as expressions, depression, and personality, face-swapping algorithms based on generative adversarial networks and diffusion models, and face generation algorithms based on text descriptions.

The facial recognition algorithm won second place in the ICPR accuracy test. The latest 200-layer neural network facial recognition algorithm achieved industry-leading accuracy of 99.27% and 99.19% in the MegaFace million-test 1:N and 1:1 tests, respectively. Its overall performance exceeded 98% in the MFR Ongoing Challenge for masked faces, also currently the highest in the industry.

The face motion unit detection algorithm won second place in the academic group of the 5th Face Motion Unit Detection Challenge at CVPR 2023!