Caffe Training stuck at 0 Iteration

[Yunhan Zhao]

I was implementing a project with Caffe but my sgd_solver stuck at 0 iterations for some reason. This is my system configuration:

System: Ubuntu 16.04

Caffe: CPU-Only

## CUDA 8.0 and cuDNN 5.1 installed 

No python/Matlab interface installed

make runtest passed without any problem.

This is the command I was running: ./build/tools/caffe train -solver models/DAN/amazon_to_webcam/solver.prototxt -weights models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel

This is my solver file:

This is what I got:

[Przemek D]

You're running CPU-only - could it be that caffe is not actually "stuck" but simply processing the first iteration very long? It is possible on CPU.
Besides, why go with CPU-only solver if you have CUDA and cuDNN installed?

[Yunhan Zhao]

I have trouble installing Caffe with CUDA 8.0 so I switched to CPU-ONLY version to run a quick test. In my case, the "CPU fan noise" significantly reduced when reaching iteration 0 so I assumed the program probably stuck in infinite loop at some point. Probably I should try again with GPU version.

[Andriy Lysak]

Had the same "issue" with CPU only. Turned out that the CPU was just very slow.

Either give it time, as in a veryyyyyyyyyy long time or use GPU