To date, BrainChip Studio utilized spiking neural networks to enable facial classification on partial faces. This partial-face mode is useful in situations where the probe image or the extracted faces may be obscured due to hats, masks, scarves or camera angle. BrainChip Studio 2018.3 uses a full-face mode to perform facial classifications. In situations where the entire face is visible in the probe image or in the extracted faces, this new mode provides a significant increase in facial classification accuracy. Depending on the dataset used, testing indicates this mode provides a 10-30 percent improvement in accuracy, without impacting throughput.
BrainChip - www.brainchip.com