0

I'm trying to run a Caffe2 model on an android device, i ran the AICameraDemo App successfully but my model init net is 7 times larger than sqeeznet model, and the load of the init net to the workspace with the predictor or with RunNetOnce takes forever.

I saw posts about this, most of them from 2018 so i'm wondering if anything change since then.

If it is not possible, is there another framework which i could run a model this big on an android device ?

Thanks for any answer.

Adam Jensen
  • 49
  • 1
  • 5
  • does your Android has strong gpu – prosti Sep 15 '19 at 11:04
  • @prosti i don't know actually what the power of the gpu but i don't think it should matter, the hole point as i understood it for me to use caffe2 is its ability to run on mobile. What the encouragement for me to use it if it can't run on mobile, i can use pytorch as a dnn framework if i want to use dnns as it is much more flexible than caffe2. – Adam Jensen Sep 16 '19 at 15:33
  • What I meant to say does some other apps run on your Android fast? – prosti Sep 16 '19 at 15:52
  • @prosti Its not the state of that art but yea, i could run almost everything on it. It a Nexus 5. The apps don't run smoothly but they run smooth enough. Youtube, netflix, chrome run fine. I tried to run the caffe2 network for 15 to 20 min so i don't think that running on a faster phone will work, i will check that but i doubt it will matter. – Adam Jensen Sep 19 '19 at 23:07

0 Answers0