This would be a crazy difficult (and fun) domain to work in. Humans are very well tuned to perceiving facial details - the margin for error would have to be tiny. I'm still uncertain if images processed this way could escape the uncanny valley, but this appears to be a much more technically feasible approach for consumer use than transparent displays, multi-cam setups, or camera arrays embedded in displays.
Re:Eye Contact in Video Conferencing (2018)
by nsb1 on Monday June 10, 2019 @08:18AM
(#20146005)
If I were a betting man, this is where I'd put my money. It seems attainable via software or hardware assisted processing once the model has been computed. All the other solutions seem too expensive or hardware intensive to me.
This would be a crazy difficult (and fun) domain to work in. Humans are very well tuned to perceiving facial details - the margin for error would have to be tiny. I'm still uncertain if images processed this way could escape the uncanny valley, but this appears to be a much more technically feasible approach for consumer use than transparent displays, multi-cam setups, or camera arrays embedded in displays.