wangxihao / smplxpp

Super fast SMPL/+H/-X implementation in C++, with CUDA support and a built-in OpenGL renderer

SMPL-X C++ implementation in Eigen and CUDA. (WIP!) AMASS integration included.

Dependencies

• Compiler supporting C++ 17 (I am using GCC 9.3.0)

Optional Dependencies

• OpenGL 3+ (Required to build the viewer)
• CUDA Toolkit with nvcc supporting C++ 14 (I have CUDA 11, not tested with earlier versions)
• Very optional: https://www.github.com/sxyu/sdf to build 'smplx-prox', which shows a cross-section of continuous SDF values inside the SMPL model

Getting the SMPL/SMPL+H/SMPL-X Models

• See data/models/README.md
• Set SMPLX_DIR environment variable to the project root directory to allow finding default data files (else you must cd to project root or in a subdirectory for the project to work)

Building

Note: I have built successfully on Ubuntu 20.04 and Windows 10 (VS2019). The project appears to run about 1.5x slower on Windows for some reason.

• To configure, mkdir build && cd build && cmake ..
  • To disable the OpenGL Viewer, replace the above cmake command with cmake .. -D SMPLX_BUILD_VIEWER=OFF
• To build, use make -j<number-of threads-here> on unix-like systems, cmake --build . --config Release else
• To install (unix only), use sudo make install (TODO: add CMake find module)

Python

To install Python bindings, use pip install . You may need to first install pybind11 from https://github.com/pybind/pybind11. Usage example:

from smplxpp import ModelS, BodyS, SequenceAMASS, Gender
# SMPL model. Use ModelX/BodyX for SMPL-X,
# ModelH/BodyH for SMPL+H, ModelXpca/BodyXPca for SMPL-X with hand PCA
model = ModelS(Gender.male)
body = BodyS(model)

seq = SequenceAMASS("/data/AMASS/KIT/10/RightTurn01_poses.npz")

# Use sequence to pose body
seq.set_shape(body)
seq.set_pose(body, 10) # Frame 10

# Do skinning (prefers GPU)
body.update()

print(body.verts) # vertices

import trimesh
tm = trimesh.Trimesh(body.vertices, model.faces) 
tm.show()

Example programs

• smplx-example: Writes SMPL-X model toout.obj
  • Usage: ./smplx-example gender where gender (optional, case insensitive) should be NEUTRAL/MALE/FEMALE; NEUTRAL is default
• smplx-viewer (if SMPLX_BUILD_VIEWER=ON in CMake): Shows an interactive 3D viewer, including parameter controls
  • Usage: `./smplx-viewer model gender device poseblends where
    • All arguments are position and optional
    • model may be S/H/X/Xp, where S means SMPL, H means SMPL+H, X means SMPL-X (with hand joints), and Xp means SMPL-X (with hand PCA). Default is S
      • New: X/Xp now specify SMPL-X v1.1 by default. Use Z/Zp for v1.0
    • gender may be NEUTRAL/MALE/FEMALE; NEUTRAL is default (case insensitive)
    • device may be gpu/cpu; gpu is default and will fallback to cpu automatically
    • poseblends may be on/off, default on; off turns off pose blendshapes, which speeds up computation dramatically
    • Example: ./smplx-viewer X MALE, ./smplx-viewer H FEMALE
    • ./smplx-viewer will open plain neutral SMPL model (if available)
• smplx-amass: AMASS viewer
  • Usage: ./smplx-amass model npz_path
    • All arguments are position and optional
    • model may be S/H/X, where S means SMPL, H means SMPL+H, X means SMPL-X. Note Xp (hand PCA) is not available for AMASS integration. Default H
    • npz_path: optionally, path to AMASS .npz to load on open
    • ./smplx-amass opens a blank viewer with option to browse for and load a npz
• smplx-sdf: View SMPL SDF

Library usage

• TBA, refer to examples (main_*.cpp) for now

License

This library is licensed under Apache v2 (non-copyleft). However, remember that the following are non-commercial research-only:

• SMPL
• SMPL-X
• Mano/SMPL+H
• AMASS dataset
• SMPLify

References

[1] SMPL: http://smpl.is.tue.mpg.de. SMPL: A Skinned Multi-Person Linear Model. Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, Michael J. Black. 2015

[2] SMPLify: http://smplify.is.tue.mpg.de. Keep it SMPL: Automatic Estimation of 3D Human Pose and Shape from a Single Image. Federica Bogo*, Angjoo Kanazawa*, Christoph Lassner, Peter Gehler, Javier Romero, Michael Black. 2016

[3] Mano: http://mano.is.tue.mpg.de. Embodied Hands: Modeling and Capturing Hands and Bodies Together. Javier Romero*, Dimitrios Tzionas*, and Michael J Black. 2017

[4] SMPL-X: http://smpl-x.is.tue.mpg.de. Expressive Body Capture: 3D Hands, Face, and Body from a Single Image. G. Pavlakos*, V. Choutas*, N. Ghorbani, T. Bolkart, A. A. A. Osman, D. Tzionas and M. J. Black. 2019

[5] AMASS: https://amass.is.tue.mpg.de/. AMASS: Archive of Motion Capture as Surface Shapes. Mahmood, Naureen and Ghorbani, Nima and Troje, Nikolaus F. and Pons-Moll, Gerard and Black, Michael J.

Vendored 3rd party libraries

The following dependencies are included in the repo and don't need to be installed

• cnpy (for npy/npz I/O) https://github.com/rogersce/cnpy
  • zlib: needed by cnpy to read npz (uses system zlib if available)
• Eigen 3.3.90 http://eigen.tuxfamily.org/
  • Note this is NEWER than the latest release! I had problems getting 3.3.7 to work with CUDA, so I vendored the version on master. You are welcome to try again.
• The following are used only if building viewer:
  • glew 2
  • glfw 3.3
  • Dear ImGui, with modifications
    • ImGui filebrowser https://github.com/AirGuanZ/imgui-filebrowser, with modifications
  • stb_image.h