can this code run on cpu-only caffe? about temporal-segment-networks HOT 3 CLOSED

It can. But it may take many days to train your networks. Running CPU-only can be achieved setting the solver to CPU mode.

from temporal-segment-networks.

thanks a lot.

from temporal-segment-networks.

Hi @yjxiong ,
I changed the solver mode to CPU. It hit a floating point exception.

*** SIGFPE (@0x7fdf7a45d45d) received by PID 26723 (TID 0x7fdf7aa3aa00) from PID 2051396701; stack trace: ***
@ 0x7fd8a21c7849 caffe::Net<>::Net()
@ 0x7fa381875c46 caffe::Solver<>::Solver()
@ 0x7f1311eb245d caffe::VideoDataLayer<>::DataLayerSetUp()
@ 0x4181f0 caffe::GetSolver<>()
@ 0x40fc2d train()
@ 0x40e5a6 main
@ 0x7fdf78b41cb0 (unknown)
@ 0x7fd8a2176a10 caffe::Solver<>::InitTrainNet()
@ 0x7f1311ef7bb5 caffe::BasePrefetchingDataLayer<>::LayerSetUp()
@ 0x7fa37fe6ef45 (unknown)
@ 0x40eb8b (unknown)
@ 0x7fdf7a45d45d caffe::VideoDataLayer<>::DataLayerSetUp()
@ 0x0 (unknown)
@ 0x7fd8a2177aae caffe::Solver<>::Init()
@ 0x7f1311fd6ab7 caffe::Net<>::Init()
@ 0x7fdf7a4a2bb5 caffe::BasePrefetchingDataLayer<>::LayerSetUp()
@ 0x7fdf7a581ab7 caffe::Net<>::Init()
@ 0x7f1311fd8849 caffe::Net<>::Net()
@ 0x7fd8a2177c46 caffe::Solver<>::Solver()
@ 0x4181f0 caffe::GetSolver<>()
@ 0x40fc2d train()
@ 0x40e5a6 main
@ 0x7fdf7a583849 caffe::Net<>::Net()
@ 0x7fdf7a532a10 caffe::Solver<>::InitTrainNet()
@ 0x7f1311f87a10 caffe::Solver<>::InitTrainNet()
@ 0x7fd8a0770f45 (unknown)
@ 0x40eb8b (unknown)
@ 0x7fdf7a533aae caffe::Solver<>::Init()
@ 0x7f1311f88aae caffe::Solver<>::Init()
@ 0x7fdf7a533c46 caffe::Solver<>::Solver()
@ 0x4181f0 caffe::GetSolver<>()
@ 0x0 (unknown)
@ 0x40fc2d train()
@ 0x40e5a6 main
@ 0x7fdf78b2cf45 (unknown)
@ 0x40eb8b (unknown)
@ 0x7f1311f88c46 caffe::Solver<>::Solver()
@ 0x4181f0 caffe::GetSolver<>()
@ 0x40fc2d train()
@ 0x40e5a6 main
@ 0x0 (unknown)
@ 0x7f1310581f45 (unknown)
@ 0x40eb8b (unknown)
@ 0x0 (unknown)

mpirun noticed that process rank 0 with PID 26722 on node saranya-H97M-D3H exited on signal 8 (Floating point exception).

@YingQian94 ,
Were you able to successfully run this on CPU-only mode?

from temporal-segment-networks.