Build textured 3D body model with two images

Example front and back images captured by Azure Kinect:

The generated texture for SMPL and SMPLX model:

The rendered result in Unity:

Requirement

• numpy
• scipy
• OpenCV
• smplify-x
• openpose

Optional requirement

• PGN

Demo

We have an exmple data in data/obj1. The front iamge is data/obj1/images/P01125-150055.jpg and the back image is data/obj1/images/P01125-150146.jpg.

To generate the texture for SMPL model, just run: python demo.py data/obj1 P01125-150055.jpg P01125-150146.jpg smpl To generate the texture for SMPLX model, just run: python demo.py data/obj1 P01125-150055.jpg P01125-150146.jpg smplx

You'll find the results in data/obj1

Using your own data

step1: prepare your image data

example can be find in ./data/obj1/images

Example image :

step2: openpose pose detection

For Linux users, you need to complie openpose following the instruction here, and for windows users, you can used the windows portable demo.
openpose.bin --display 0 --render_pose 1 --image_dir ./data/obj1/images --write_json ./data/obj1/keypoints --write_images ./data/obj1/pose_images --hand --face

Example image :

step3: fit smpl/smplx model

Please follow the instruction here
An example command is: python smplifyx/main.py --config cfg_files/fit_smplx.yaml --data_folder ./data/obj1 --output_folder ./data/obj1/smplx --visualize=True --model_folder models --vposer_ckpt vposer_v1_0
Please copy the output data to ./data/obj1/smplx or ./data/obj1/smpl

Example image :

step4(optional): get PGN segmentation

Please follow the instruction here
Please copy the output data to ./data/obj1/PGN

Example image :

step4: texture generation

run python demo.py data_path front_img back_img smplx

Example image :