expressions updated to fit libtorch v1.4.0 and opencv4.0

[Ageliss]

Thanks for sharing the code to the cpp libtorch starters. I found something needs to be updated and hope others won't need to do it again.

1. prediction.cpp should be updated as pull request.

2. model_trace.py: "model = torchvision.models.resnet18(pretrained=True)" may result "SSL: CERTIFICATE_VERIFY_FAILED", which can be solved by adjusting "https://" into "http://" in "torchvision.models.resnet.model_urls"...

3. the installation of dependant libs of opencv and libtorch maybe hard for starters, will be better if mentioned in step 0 or 1 in README.md
   like:
   install libs of opencv:
   for MacOS, xcode-select --install, install XCODE in app store(will automatically install cmake make gcc g++ I guess), apt install opencv(will install opencv4.0 and take a really long time).....
   install libtorch:
   wget https://download.pytorch.org/libtorch/nightly/cpu/libtorch-shared-with-deps-latest.zip
   unzip libtorch-shared-with-deps-latest.zip

for Linux:
apt install gcc
apt install c++
apt install cmake
apt install make
and etc...

4. when doing "make", some linux machines with RAM less than 1.5G(maybe) will get internal compiler error. So recommend starters to compile on the bigger RAM machine...

[BIGBALLON]

Hi, @Ageliss There are lots of changes after PyTorch tagv1.0. , thanks for your PR!!! It's very helpful for other users