MotionProcessTools

useful tools for motion data processing, related to SMPL, BVH, FBX, etc.

SetUP

python==3.11

pip install -r requirements.txt

prepare the SMPL model

follow the motionGPT to prepare the smpl model, you should set them as the following structure.

📦deps
┗ 📂smpl
┃ ┗ 📂smpl_models
┃ ┃ ┣ 📂smpl
┃ ┃ ┃ ┣ 📜readme.txt
┃ ┃ ┃ ┣ 📜SMPL_FEMALE.pkl
┃ ┃ ┃ ┣ 📜SMPL_MALE.pkl
┃ ┃ ┃ ┗ 📜SMPL_NEUTRAL.pkl
┃ ┃ ┣ 📜gmm_08.pkl
┃ ┃ ┣ 📜neutral_smpl_mean_params.h5
┃ ┃ ┣ 📜smpl.faces
┃ ┃ ┣ 📜smpl.tar.gz
┃ ┃ ┣ 📜smplx_parts_segm.pkl
┃ ┃ ┗ 📜SMPL_downsample_index.pkl

Usage

1.Kalman_Filter for SMPL-like joints npy data

if your mocap data is noisy, you can use Kalman_Filter to smooth the data.

this way is notice: the data should be transfor to SMPL joints.

python ./tools/Kalman_Filter.py

2. SMPL Joints to BVH

modify the setting in joints2bvh.py, then run the script.

python joints2bvh.py

based on animationGPT for this part.

trans joints[fnum, 24, 3] to smpl pose parameters

3. SMPL Joints to pose param θ

3.1 using symplify3d

this part comes form motionGPT

python -m Smplify3D.fit --dir YOUR_JOINTS_FOLDER --save_folder DIR_TO_SAVE --cuda 0

this will fit joints to smply, and generate the SMPLPose.pkl which is for bvh; and generate mesh.npy in the same folder in YOUR_JOINTS_FOLDER, which is for render;

3.2 using naive-HyberIK

python -m HyberIK.fit -- dir YOUR_JOINTS_FOLDER --save_folder DIR_TO_SAVE --cuda 0

4. SMPL pose param θ to bvh

once you get the SMPLPose.pkl, you can use the following script to generate bvh file.

python smpl2bvh.py --gender MALE --poses ${PATH_TO_Y0UR_INPUT} --fps 60 --output ${PATH_TO_SAVE} --mirror

you can find more details on this smpl2bvh

Smplify3D(left) and HyberIK(right)

5. BVH to FBX

you need to install blender at first

Path_to_your_blender -b --python ./tools/bvh2fbx.py -- --BVHF BVH_FOLDER_PATH [--FBXF FBX_FOLDER_PATH]

you can use fbx in unity and UE to make animation now :) enjoy it.

Related

• animationGPT
• AvatarSoul
• motionGPT
• smpl2bvh

thanks for their great work.