“…Researchers have used a variety of cameras, from infrared cameras [ 14 ] and high-quality monochrome cameras equipped with special filters [ 15 , 16 , 17 , 18 ] to off-the-shelf webcams [ 19 , 20 , 21 , 22 , 23 ], to estimate SpO 2 by capturing subtle light intensity changes on the face. Deep learning techniques have achieved state-of-the-art performance for the remote measurement of physiological signs such as HR [ 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 ] and RR [ 36 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 ]. However, remote SpO 2 measurement is still in its infancy, with only a few papers using convolutional neural networks (CNNs) to predict SpO 2 from RGB facial videos [ 48 , 49 , 50 ].…”