A 100% compatiable C++ implemention for python package smplx. Easy to use and integrate into any pytorch/libtorch workflow. Original SMPL project can be found on this website

See here for more instructions.

1. if you are using SMPL, you should convert SMPL_*.pkl to .npz format by using provided python script.

python pkl2npz.py /path/to/SMPL_*.pkl ...

• libtorch provides tensor computation and network definition. if you prefer running on Nvidia GPU, please download CUDA version liborch you can choose libtorch version according to your environgment from https://pytorch.org/get-started/locally/
• zlib for read .npz file

apt install zlib1g-dev
# or you can install by conda
conda install zlib

• compiler supports C++17 standards
  • mingw64 on Windows not supported due to incompatiable ABI
  • vc141, vc142, vc143 are tested (correspoding to Visual Studio 2017/2019/2022)
  • gcc > 8.0 tested
  • clang > 5.0 in theory (not tested yet)

If you have installed libtorch and zlib in other paths, please modify corresponding path in commands below

Make sure your current work directory is project root. CUDA version

g++ -o a.out -std=c++17 <your_source_file_here> \
body_models.cpp joint_names.cpp lbs.cpp npyio.cpp \
vertex_ids.cpp vertex_joint_selector.cpp -Iinclude\
-Ilibtorch/include -Ilibtorch/include/torchcsrc/api/include -Izlib/include \
-Llibtorch/lib -Lzlib/lib \
-lz -lc10  -lc10_cuda -ltorch -ltorch_cpu -ltorch_cuda

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:libtorch/lib:zlib/lib

CPU version

g++ -o a.out -std=c++17 <your_source_file_here> \
body_models.cpp joint_names.cpp lbs.cpp npyio.cpp \
vertex_ids.cpp vertex_joint_selector.cpp -Iinclude\
-Ilibtorch/include -Ilibtorch/include/torchcsrc/api/include -Izlib/include \
-Llibtorch/lib -Lzlib/lib \
-lz -lc10 -ltorch -ltorch_cpu

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:libtorch/lib:zlib/lib

Modify CMakeLists.txt and change the TORCH_DIR to /path/to/your/libtorch/share/cmake/Torch.

mkdir build && cd build
cmake ../
make

It has the same api with python package smplx.