To build high-performance inference using OpenPose-Plus, the following dependencies should be installed on your machine:
- OpenCV3.4+
- GFlags
- TensorRT(version 5 & 7 are tested, other versions may also work)
- Ubuntu:
sudo apt-get install libgflags-dev
Please follow the instructions here.
Make sure that the header files & libraries of TensorRT & CUDA are in the /usr/local/cuda/targets/${ARCH}-linux
's include
& lib
folder.
Or you may need to set CUDA_RT
as /path/to/TensorRT
when executing cmake commands.
# Tools we need when building.
sudo apt-get install subversion
make pack # If you want to enable profiling tracing, use `make pack_trace`
./scripts/download-test-data.sh
./scripts/download-pretrained-inf-models.sh
cd cmake-build/Linux
./example-batch-detector
# Or `./example-batch-detector --use_f16=true` if your GPU support float16
Then you'll see the output images in the cmake-build/Linux
folder.
If you cannot execute the binary file successfully, you can refer to the tensorrt.log
for more details.
If you want to share your results on your machine, welcome to PR!
- FP32: ~50 FPS