EasyMocap Demo 安装各种错误的坑

安装环境与依赖

OpenPose on Linux Mint 21.3(Ubuntu 22.04)

https://github.com/AClon314/tauri-vuetify-learn/blob/c57c203a27eadd4a39142002fb61230bf750d6d4/src-tauri/capabilities/mobile.json#L31C5-L38C7
装好了pytorch，cuda, cuDnn，还需要手动编译Openpose，卡在了make -j8，一些error和一堆warning

/usr/local/cuda/include/cuda/std/detail/libcxx/include/__cuda/ptx/instructions/../ptx_helper_functions.h:60:44: error: ‘__cvta_generic_to_global’ was not declared in this scope
   60 |   return static_cast<_CUDA_VSTD::uint64_t>(__cvta_generic_to_global(__ptr));
      |                                            ^~~~~~~~~~~~~~~~~~~~~~~~
/usr/local/cuda/include/cuda/std/detail/libcxx/include/__cuda/ptx/instructions/../ptx_helper_functions.h: In function ‘_Tp* cuda::ptx::__4::__from_ptr_smem(size_t)’:
/usr/local/cuda/include/cuda/std/detail/libcxx/include/__cuda/ptx/instructions/../ptx_helper_functions.h:73:33: error: there are no arguments to ‘__cvta_shared_to_generic’ that depend on a template parameter, so a declaration of ‘__cvta_shared_to_generic’ must be available [-fpermissive]
   73 |   return reinterpret_cast<_Tp*>(__cvta_shared_to_generic(__ptr));
      |                                 ^~~~~~~~~~~~~~~~~~~~~~~~
/usr/local/cuda/include/cuda/std/detail/libcxx/include/__cuda/ptx/instructions/../ptx_helper_functions.h:73:33: note: (if you use ‘-fpermissive’, G++ will accept your code, but allowing the use of an undeclared name is deprecated)
/usr/local/cuda/include/cuda/std/detail/libcxx/include/__cuda/ptx/instructions/../ptx_helper_functions.h: In function ‘_Tp* cuda::ptx::__4::__from_ptr_gmem(size_t)’:
/usr/local/cuda/include/cuda/std/detail/libcxx/include/__cuda/ptx/instructions/../ptx_helper_functions.h:94:33: error: there are no arguments to ‘__cvta_global_to_generic’ that depend on a template parameter, so a declaration of ‘__cvta_global_to_generic’ must be available [-fpermissive]
   94 |   return reinterpret_cast<_Tp*>(__cvta_global_to_generic(__ptr));
      |                                 ^~~~~~~~~~~~~~~~~~~~~~~~
In file included from /usr/local/cuda/include/thrust/system/cuda/detail/util.h:48,

考虑手动换openpose为mediapipe
https://github.com/Totemi1324/caffe-installation-guide

更换docker

仍然遇到caffe编译失败的问题

docker run -d -p 8080:80 -m 4g --name CONTAINER_NAME --hostname CONTAINER_INNER_HOSTNAME -v /home/user/data:/data -e VAR=1 --rm my_image

-d: 后台运行
-v: 挂载host的目录到guest目录
-e: 环境变量
--rm: 退出后自动删除，一次性容器
-m: 内存限制

在容器中使用gpu：https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html
rootless无根模式下，容器使用gpu：https://github.com/NVIDIA/nvidia-docker/issues/1155

最后发现不如用podman，docker竟然还分rootless和root模式