I've converted the SqueezeNet model from caffe to snpe and it works well. But when I tried to continually execute the model on my Nexus 6 CPU, the speed is quite weird. The first two rounds always cost around 80ms, and then every round costs 170ms ~ 200ms. I repeated several times and the results are similar. What is the reason for that? Appreciate your replies.
Inconsistent speed when running SqueezeNet
Posted: Mon, 2017-12-18 23:58