Support pytorch 0.4 and support cpu during test or demo

The code has been updated to pytorch 0.4.

Thanks to the .to(device), now the model can be simply change between cpu and cuda. Defaulty, if there is no cuda device, when you run test or demo, the model will automatically run under cpu.

Version 1.0

The current version is stable but not as good as tf-faster-rcnn master.

There will be a push to the master. Just to backup here.