Summary: | In recent years, the research community has developed and proposed various face anti-spoofing models that involves multiple modalities. The demand for these methods continue to rise due to advancement of sensors and increasing usage of biometric security. However, there has been no research on compressed multi-modality face anti-spoofing methods that can offer good generalization performance. This project proposes a compressed multi-modality face anti-spoofing model based on an existing state-of-the-art method. The proposed model requires a lower amount of computational resource and has a much shorter inference time, suitable for deployment to edge devices. It manages to obtain comparable performance to that of the state-of-the-art face anti-spoofing model.
|