【2020年版】NVIDIA Jetson Nano対応の OpenPoseをビルドする手順

【2020年版】NVIDIA Jetson Nano、Jetson Xavier NXの便利スクリプト

cudnn.h not found

-- Building with CUDA.
-- CUDA detected: 10.2
-- Found cuDNN: ver. ??? found (include: /usr/include, library: /usr/lib/aarch64-linux-gnu/libcudnn.so)
CMake Error at cmake/Cuda.cmake:263 (message):
  cuDNN version >3 is required.
Call Stack (most recent call first):
  cmake/Cuda.cmake:291 (detect_cuDNN)
  CMakeLists.txt:417 (include)

【2020年版】NVIDIA JetPack 4.4 DP Developer Previewで OpenPoseのビルドエラーの対策方法

● JetPack 4.3
jetson@jetson-desktop:~ $ ls -l /usr/include/cudnn.h
lrwxrwxrwx 1 root root 26 12月 17 03:53 /usr/include/cudnn.h -> /etc/alternatives/libcudnn

jetson@jetson-desktop:~ $ ls -l /usr/lib/aarch64-linux-gnu/libcudnn.so
lrwxrwxrwx 1 root root 29 12月 17 03:53 /usr/lib/aarch64-linux-gnu/libcudnn.so -> /etc/alternatives/libcudnn_so

jetson@jetson-desktop:~ $ uname -a
Linux jetson-desktop 4.9.140-tegra #1 SMP PREEMPT Mon Dec 9 22:47:42 PST 2019 aarch64 aarch64 aarch64 GNU/Linux

jetson@jetson-desktop:~ $ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 18.04.3 LTS
Release:        18.04
Codename:       bionic

【2020年版】NVIDIA Jetson Nanoの初期設定、最高速で動かす設定、空きメモリを増やす方法等

【2020年版】NVIDIA Jetson用に最新の CMake 3.17.3をビルドしてインストールする方法

# CMake v3.17.3のビルド
# OpenPoseのビルドには最新の cmakeが必要なので cmakeをビルドする
# 上の記事を参照

# お決まりの sudo apt-get updateで最新状態に更新する
sudo apt-get update

# cmakeは入れない事！
sudo apt-get -y install git

# user@user-desktop:~$ cmake --version
# cmake version 3.10.2
# apt-get installの cmakeのバージョンは OpenPoseのビルドでエラーが出る
# CMake Error: The following variables are used in this project, but they are set to NOTFOUND.

# cmakeのバージョン確認
cmake --version
# cmake version 3.17.3
# CMake suite maintained and supported by Kitware (kitware.com/cmake).

# OpenPoseのビルド

# OpenPoseのソースリストをダウンロードする
# build OpenPose
cd
# OpenPose v1.6.0 Apr 27, 2020
git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose -b v1.6.0 --depth 1
cd openpose
sudo bash ./scripts/ubuntu/install_deps.sh
# Collecting opencv-python
#  Could not find a version that satisfies the requirement opencv-python (from versions: )
# No matching distribution found for opencv-python
# ↑これは気にしない

mkdir build
cd build

# cmake ..
# CMake Warning:
#   Manually-specified variables were not used by the project:
#     CUDA_ARCH_BIN

cmake .. \
  -D CMAKE_BUILD_TYPE=Release \
  -D CUDA_ARCH_BIN="5.3"

# NVIDIA CUDA GPUs Compute Capability
# https://developer.nvidia.com/cuda-gpus
# Compute Capability Jetson Nano = 5.3
# -D CUDA_ARCH_BIN="5.3"
# Compute Capability Jetson Xavier = 7.2
# -D CUDA_ARCH_BIN="7.2"

# -D CUDA_ARCH_BIN="5.3" -D CUDA_ARCH_PTX="5.3"
# WITH_CUDA WITH_CUBLAS CUDA_GENERATION
# -- Found CUDA: /usr/local/cuda (found version "10.0")
# -- Building with CUDA.
# -- CUDA detected: 10.0
# -- Found cuDNN: ver. 7.6.3 found (include: /usr/include, library: /usr/lib/aarch64-linux-gnu/libcudnn.so)
# -- Added CUDA NVCC flags for: sm_53
# -- Found cuDNN: ver. 7.6.3 found (include: /usr/include, library: /usr/lib/aarch64-linux-gnu/libcudnn.so)

# -- Found OpenCV: /usr (found version "4.1.1")

# 4コアでビルドで時間短縮
make clean
time make -j4
# real    27m47.666s
# user    82m48.020s
# sys     5m16.240s

sudo make install

# 動画を mp4形式で書き出す場合に FFmpegが必要
sudo apt-get -y install ffmpeg

【2020年版】NVIDIA Jetson Nano対応の FFmpegをビルドする手順

free -h
#               total        used        free      shared  buff/cache   available
# Mem:           3.9G        283M        2.0G         28M        1.6G        3.4G
# Swap:          1.9G          0B        1.9G
# Jetson Nanoの Jetpack 4.3で解像度を無指定で動いた

cd
cd openpose

# 動画 mp4
./build/examples/openpose/openpose.bin --video examples/media/video.avi --display 0 --model_folder ./models --write_video output_video.mp4
# OpenPose demo successfully finished. Total time: 414.525283 seconds. Nano Jetpack 4.3
# OpenPose demo successfully finished. Total time: 152.889621 seconds. Xavier NX Jetpack 4.4 DP
ls -l output_video.mp4
# -rw-rw-r-- 1 jetson jetson 1918270  6月 16 21:04 output_video.mp4

# Jetson Nanoの Jetpack 4.3は無指定でも動いたが、Jetpack 4.4 DPは Killedで失敗した
# Jetson Nanoの Jetpack 4.3でも Killedで失敗する場合がある（丁度ギリギリ？）
# Killedで失敗する時はメモリ不足なので
# --net_resolution 480x256で認識の解像度を落とすと動く
./build/examples/openpose/openpose.bin --video examples/media/video.avi --display 0 --model_folder ./models --write_video output_video.mp4 --net_resolution 480x256

# -fps_max 15でフレームレートを指定できる
# -disable_blendingで骨格だけになる（元絵を合成しない）
./build/examples/openpose/openpose.bin --video examples/media/video.avi --display 0 --model_folder ./models --write_video output_video.mp4 --net_resolution 480x256 -fps_max 15 -disable_blending

cd
cd openpose

# 動画 avi
# time ./build/examples/openpose/openpose.bin --video examples/media/video.avi --display 0 --model_folder ./models --write_video output_video.avi --net_resolution 320x176 -fps_max 15
time ./build/examples/openpose/openpose.bin --video examples/media/video.avi --display 0 --model_folder ./models --write_video output_video.avi
ls -l *.avi
# -rw-rw-r-- 1 jetson jetson 17226188  6月 16 22:52 output_video.avi

cd
cd openpose

# 静止画
./build/examples/openpose/openpose.bin -image_dir ./examples/media/ --display 0 --model_folder ./models --write_images ./output/
# OpenPose demo successfully finished. Total time: 55.865296 seconds. Nano Jetpack 4.3
# OpenPose demo successfully finished. Total time: 35.759475 seconds. Xavier NX Jetpack 4.4 DP
ls -l ./examples/media/
ls -l ./output/

# Jetson Nanoの Jetpack 4.3は無指定でも動いたが、Jetpack 4.4 DPは Killedで失敗した
# Jetson Nanoの Jetpack 4.3でも Killedで失敗する場合がある（丁度ギリギリ？）
# Killedで失敗する時はメモリ不足なので
# --net_resolution 480x256で認識の解像度を落とすと動く
./build/examples/openpose/openpose.bin -image_dir ./examples/media/ --display 0 --model_folder ./models --write_images ./output/ --net_resolution 480x256

    const std::string imageToVideoCommand = "ffmpeg -y -framerate " + std::to_string(upImpl->mFps)
        + " -i '" + upImpl->mTempImageFolder + "/%12d_rendered.jpg'"
        + " -c:v libx264 -pix_fmt yuv420p '"
        + upImpl->mVideoSaverPath + "'";

NVIDIA Jetson Nanoで OpenCV 3をビルドしてインストールする方法、NVCaffe等の OpenCV 4未対応を動かす

【ビルド版】NVIDIA Jetson Nanoで NVIDIA版 Caffe NVCaffeを GPUパワーで動かしてキモイ絵をモリモリ量産

# NVCaffe May 1, 2019 v0.17.3 fd6cf7a
# NVCaffeをビルドする
# CAFFE_HOME等の環境変数を設定する

echo ${CAFFE_HOME}
# /home/jetson/nvcaffe

# OpenPoseのビルドに必要なファイルをコピーする
# fatal error half_float/half.hpp No such file or directory
cp -r ${CAFFE_HOME}/3rdparty/* ${CAFFE_HOME}/include

# fatal error caffe/proto/caffe.pb.h No such file or directory
cd ${CAFFE_HOME}
protoc src/caffe/proto/caffe.proto --cpp_out=.
mkdir include/caffe/proto
mv src/caffe/proto/caffe.pb.h include/caffe/proto

CAFFE_PATH=${CAFFE_HOME}

echo ${CAFFE_PATH}

# ls -l ${CAFFE_PATH}/build/lib/
# libcaffe-nv.so

Caffe_INCLUDE_DIRS=${CAFFE_PATH}/include
echo ${Caffe_INCLUDE_DIRS}
# /home/jetson/nvcaffe/include

Caffe_LIBS=${CAFFE_PATH}/build/lib/libcaffe-nv.so
echo ${Caffe_LIBS}
# /home/jetson/nvcaffe/build/lib/libcaffe-nv.so

ls -l ${CAFFE_PATH}/include
ls -l ${CAFFE_PATH}/build/lib/libcaffe-nv.so

# OpenCV 3.4.10
ls -l /usr/local/share/OpenCV

# OpenPose
cd
git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose -b v1.6.0 --depth 1
cd openpose

# undefined reference to symbol '_ZN5boost6system15system_categoryEv'
sed -i 's/resizeTest.cpp)/)/' ./examples/tests/CMakeLists.txt
cat ./examples/tests/CMakeLists.txt | grep resizeTest

mkdir build
cd build

# Caffe_LIBSは間違い。Caffe_LIBS_RELEASEが正しい。ただし、Caffe_LIBSも必要。
# -D CUDA_ARCH_BIN="5.3"は不要
# -D CMAKE_BUILD_TYPE=Release
# https://cmake.org/cmake/help/v3.17/variable/CMAKE_BUILD_TYPE.html
# CMAKE_BUILD_TYPE Possible values are empty, Debug, Release, RelWithDebInfo, MinSizeRel
cmake .. \
  -D CMAKE_BUILD_TYPE=Release \
  \
  -D DL_FRAMEWORK=NV_CAFFE \
  -D BUILD_CAFFE=OFF \
  -D Caffe_INCLUDE_DIRS=${CAFFE_PATH}/include \
  -D Caffe_LIBS=${CAFFE_PATH}/build/lib/libcaffe-nv.so \
  -D Caffe_LIBS_RELEASE=${CAFFE_PATH}/build/lib/libcaffe-nv.so \
  \
  -D OpenCV_DIR=/usr/local/share/OpenCV \
  \
  -D DOWNLOAD_BODY_COCO_MODEL=ON \
  -D DOWNLOAD_BODY_MPI_MODEL=ON

# COCOと MPIのモデルデータのダウンロード無し
cmake .. \
  -D CMAKE_BUILD_TYPE=Release \
  \
  -D DL_FRAMEWORK=NV_CAFFE \
  -D BUILD_CAFFE=OFF \
  -D Caffe_INCLUDE_DIRS=${CAFFE_PATH}/include \
  -D Caffe_LIBS=${CAFFE_PATH}/build/lib/libcaffe-nv.so \
  -D Caffe_LIBS_RELEASE=${CAFFE_PATH}/build/lib/libcaffe-nv.so \
  \
  -D OpenCV_DIR=/usr/local/share/OpenCV

CMake Error at CMakeLists.txt:839 (message):
  Caffe not found.  Either turn on the BUILD_CAFFE option or specify the path
  of Caffe includes
          and libs using -DCaffe_INCLUDE_DIRS and -DCaffe_LIBS.

# Caffe_LIBSは間違い。Caffe_LIBS_RELEASEが正しい。
# CMU-Perceptual-Computing-Lab/openpose/CMakeLists.txt
# # Check if the user specified caffe paths
# if (Caffe_INCLUDE_DIRS AND Caffe_LIBS_RELEASE AND NOT BUILD_CAFFE)
  -D Caffe_INCLUDE_DIRS=${CAFFE_PATH}/include/caffe \
  -D Caffe_LIBS_RELEASE=${CAFFE_PATH}/build/lib/libcaffe-nv.so \

# ただし Caffe_LIBSもライブラリの探索に使っているので必要
# cmake/Modules/FindCaffe.cmake
# find_library(Caffe_LIBS NAMES caffe

[  0%] Building CXX object src/openpose/CMakeFiles/openpose.dir/core/arrayCpuGpu.cpp.o
/home/jetson/openpose/src/openpose/core/arrayCpuGpu.cpp:3:14: fatal error: caffe/blob.hpp: No such file or directory
     #include <caffe/blob.hpp>
              ^~~~~~~~~~~~~~~~
compilation terminated.
src/openpose/core/CMakeFiles/openpose_core.dir/build.make:95: recipe for target 'src/openpose/core/CMakeFiles/openpose_core.dir/arrayCpuGpu.cpp.o' failed
make[2]: *** [src/openpose/core/CMakeFiles/openpose_core.dir/arrayCpuGpu.cpp.o] Error 1

 Wrong
  -D Caffe_INCLUDE_DIRS=${CAFFE_PATH}/include/caffe
 Correct
  -D Caffe_INCLUDE_DIRS=${CAFFE_PATH}/include

[  6%] Building CXX object src/openpose/CMakeFiles/openpose.dir/core/arrayCpuGpu.cpp.o
In file included from /home/jetson/openpose/3rdparty/caffe/include/caffe/common.hpp:32:0,
                 from /home/jetson/openpose/3rdparty/caffe/include/caffe/blob.hpp:11,
                 from /home/jetson/openpose/src/openpose/core/arrayCpuGpu.cpp:3:
/home/jetson/openpose/3rdparty/caffe/include/caffe/util/float16.hpp:12:10: fatal error: half_float/half.hpp: No such file or directory
 #include "half_float/half.hpp"
          ^~~~~~~~~~~~~~~~~~~~~
compilation terminated.
src/openpose/CMakeFiles/openpose.dir/build.make:249: recipe for target 'src/openpose/CMakeFiles/openpose.dir/core/arrayCpuGpu.cpp.o' failed
make[2]: *** [src/openpose/CMakeFiles/openpose.dir/core/arrayCpuGpu.cpp.o] Error 1

${CAFFE_PATH}/3rdparty
　の中身を
${CAFFE_PATH}/include
　にコピーする。

cp -r ${CAFFE_PATH}/3rdparty/* ${CAFFE_PATH}/include

$ ls -l ${CAFFE_PATH}/include
drwxrwxr-x 5 jetson jetson 4096  6月 22 21:59 caffe
drwxrwxr-x 3 jetson jetson 4096  6月 23 23:07 cub
drwxrwxr-x 2 jetson jetson 4096  6月 23 23:07 half_float

[  1%] Building CXX object src/openpose/CMakeFiles/openpose.dir/core/arrayCpuGpu.cpp.o
In file included from /home/jetson/openpose/3rdparty/caffe/include/caffe/blob.hpp:12:0,
                 from /home/jetson/openpose/src/openpose/core/arrayCpuGpu.cpp:3:
/home/jetson/openpose/3rdparty/caffe/include/caffe/syncedmem.hpp:8:10: fatal error: caffe/proto/caffe.pb.h: No such file or directory
 #include "caffe/proto/caffe.pb.h"
          ^~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
src/openpose/core/CMakeFiles/openpose_core.dir/build.make:95: recipe for target 'src/openpose/core/CMakeFiles/openpose_core.dir/arrayCpuGpu.cpp.o' failed
make[2]: *** [src/openpose/core/CMakeFiles/openpose_core.dir/arrayCpuGpu.cpp.o] Error 1

# cannot find #include "caffe/proto/caffe.pb.h" #1761
# https://github.com/BVLC/caffe/issues/1761
# In the directory you installed Caffe to
cd ${CAFFE_PATH}
protoc src/caffe/proto/caffe.proto --cpp_out=.
mkdir include/caffe/proto
mv src/caffe/proto/caffe.pb.h include/caffe/proto

Scanning dependencies of target openpose_calibration
[ 16%] Building CXX object src/openpose/calibration/CMakeFiles/openpose_calibration.dir/cameraParameterEstimation.cpp.o
make[2]: *** No rule to make target 'caffe-NOTFOUND', needed by 'src/openpose/3d/libopenpose_3d.so'.  Stop.
CMakeFiles/Makefile2:745: recipe for target 'src/openpose/3d/CMakeFiles/openpose_3d.dir/all' failed
make[1]: *** [src/openpose/3d/CMakeFiles/openpose_3d.dir/all] Error 2

エラー多すぎ！！ ひどい！！ ひどすぎる借金！！ ハイドロジェン！！
-D Caffe_LIBS=${CAFFE_PATH}/build/lib/libcaffe-nv.so
　が必要

[ 97%] Linking CXX executable resizeTest.bin
/usr/bin/ld: CMakeFiles/resizeTest.bin.dir/resizeTest.cpp.o: undefined reference to symbol '_ZN5boost6system15system_categoryEv'
//usr/lib/aarch64-linux-gnu/libboost_system.so.1.65.1: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
examples/tests/CMakeFiles/resizeTest.bin.dir/build.make:140: recipe for target 'examples/tests/resizeTest.bin' failed
make[2]: *** [examples/tests/resizeTest.bin] Error 1
CMakeFiles/Makefile2:1813: recipe for target 'examples/tests/CMakeFiles/resizeTest.bin.dir/all' failed

死ぬる。。。

sudo apt-get install libboost-all-dev
# libboost-all-dev is already the newest version (1.65.1.0ubuntu1).

sudo apt-get install libgflags-dev

CMakeLists.txt
-lboost_system

DSO missing from command line
-lpthread

　対処方法がわからないので諦めて resizeTest.cppのビルドを除外する。
nano ../examples/tests/CMakeLists.txt
set(EXAMPLE_FILES
    handFromJsonTest.cpp
#    resizeTest.cpp)
    )

#include <cstdio>
int main()
{
  int count = 0;
  if (cudaSuccess != cudaGetDeviceCount(&count)) return -1;
  if (count == 0) return -1;
  for (int device = 0; device < count; ++device)
  {
    cudaDeviceProp prop;
    if (cudaSuccess == cudaGetDeviceProperties(&prop, device))
      std::printf("%d.%d ", prop.major, prop.minor);
  }
  return 0;
}

nvcc detect_cuda_archs.cu
./a.out
# 5.3 Jetson Nano

【2020年版】NVIDIA Jetson Nano、Jetson Xavier NXの便利スクリプト

【2020年】Jetson Xavier NX 開発者キットが安かったので衝動買いした件、標準販売価格５万円が４万４千円！

【2020年版】NVIDIA Jetson用に最新の CMake 3.17.3をビルドしてインストールする方法

【2020年版】NVIDIA JetPack 4.4 DP Developer Previewで OpenPoseのビルドエラーの対策方法

Jetson Nano、Jetson Xavier NX 開発者キット TensorFlow on Jetson Platform

NVIDIA Jetson Nano 開発者キットを買ってみた。メモリ容量 4GB LPDDR4 RAM

【2020年版】NVIDIA Jetson Nanoの初期設定、最高速で動かす設定、空きメモリを増やす方法等

【2020年版】NVIDIA Jetson Nano JetPackのバージョン情報まとめ、JetPack 4.4は仕様変更の影響が大きい

【2020年版】NVIDIA Jetson Nano対応の FFmpegをビルドする手順

NVIDIA Jetson Nanoで OpenCV 3をビルドしてインストールする方法、NVCaffe等の OpenCV 4未対応を動かす

NVIDIA Jetson Nanoで Visual Studio Code Open Sourceをビルドして実行する

【2020年版】NVIDIA Jetson Nanoで TensorFlowの StyleGANを動かして、顔画像を生成

【2020年版】NVIDIA Jetson Nanoで StyleGANの改良版の StyleGAN2で自然な画像を生成

【2020年版】Jetson Nanoで StyleGANを動かして可愛い美少女のアニメ顔を大量生産する方法

【メモリ8GB】Raspberry Pi 4 Model B 8GBを KSYで最安値で購入。ベンチマークレビュー

【Vulkan】Raspberry Pi 4 Model Bで Vulkanドライバをビルドして 3Dグラフィックのデモを動かす

PIP機能付きの 4K対応の 4入力 1出力の HDMIセレクターを買ってみた、HDMI機器が複数有る場合に便利

EDID保持機能付きの 4K対応の 4入力 2出力の マトリックス切り替え HDMIセレクター、液晶画面 2台と使用で最強