FlexDNN: Input-Adaptive On-Device Deep Learning for Efficient Mobile Vision | IEEE Conference Publication | IEEE Xplore