[Pose Estimation] Openpose 설치 및 demo 실행

jupyter-notebook --ip=0.0.0.0 --no-browser --port=8897

---

CUDA 버전 확인
$ nvcc --version

이 전에 설치한 openpose 삭제
$ rm -rf 'openpose'

---

nvcc fatal unsupported gpu architecture 'compute_75' 에러 발생...
아래와 같이 nvcc compiler가 더 낮은 버전의 compute capability를 사용하도록 임의로 조정해주면 해결 된다.
$ export TORCH_CUDA_ARCH_LIST="7.0"

---

$ git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose.git
$ cd openpose
$ git submodule init
$ git submodule update

$ cp scripts/ubuntu_deprecated/Makefile.config.Ubuntu16_cuda7.example Makefile.config
$ cp 3rdparty/caffe/Makefile.config.Ubuntu16_cuda7.example 3rdparty/caffe/Makefile.config

$ mkdir build && cd build
$ cmake ..

$ make -j`nproc`
$ (sudo) make install

---

# Only Body(비디오 입력)
$ cd ..
$ ./build/examples/openpose/openpose.bin --video examples/media/video.avi

# With face and hands
$ ./build/examples/openpose/openpose.bin --video examples/media/video.avi --face --hand

# openpose/data/사진들 있을경우
$ ./build/examples/openpose/openpose.bin --image_dir data/ -write_images data/

# 카메라 입력을 통한 실행 (리언센스의 RGB 카메라 Index는 2)
$ ./build/examples/openpose/openpose.bin --camera=2

# Face, Hand Pose 모델 추가
$ ./build/examples/openpose/openpose.bin --camera=2 --face --hand