How to accelerate the inference speed by using inference engine (ex:Intel Openvino, NVIDIA TensorRT) instead of using caffe?

[jackgao0323]

Type of Issue

• Help wanted

Your System Configuration

2. OpenPose version: Latest GitHub code

3. General configuration:

   • Installation mode: CMake
   • Operating system : Windows 10
   • Release or Debug mode : Release
   • Compiler : VS2015 community

4. Non-default settings:

   • 3-D Reconstruction module added? : no
   • Any other custom CMake configuration with respect to the default version? : no

5. 3rd-party software:

   • Caffe version: Default from OpenPose
   • CMake version : CMake 3.8.0
   • OpenCV version: OpenPose default

6. If Windows system:

   • Portable demo or compiled library?

I want to accelerate the inference speed by using inference engine (ex:Intel Openvino, NVIDIA TensorRT) instead of using caffe.

I want to know how to get the input of the caffe model in your program and what should I do after I get the output?

Thank you very much.

[bushibushi]

Hello, I started a TensorRT PR last year but left it after the big PIMPL refactoring messed things up just when I got less time to work on this repo. You can still see it in unmerged PRs, it can give you ideas on how to make it work for actual repo configuration. I only got to replace the main NN inference part, did a few computation changes for other ops but didn't replace some of the caffe components.

[jackgao0323]

@bushibushi Thank you very much. I'll take a look right now.

[bushibushi]

I'd suggest checking out early in the branch where i am certain things worked like 8023fb1