rhythmjnh/tf_smpl

SMPLify, Keep it SMPL, Estimation the pose and shape, TensorFlow.

TF_SMPL

TF_SMPL is a TensorFlow implementation of Keep it SMPL: Automatic Estimation of 3D Human Pose and Shape from a Single Image, it can estimate the 3D pose and shape of the human body from a single image.

Dependencies

This code uses Python 2.7 and need the following dependencies, I recommend you to use a virtual environment.

• numpy, tensorflow-cpu, matplotlib, tqdm, time, cv2, pickle, opendr(if you want to render the result), scipy.

Usage

Download the model, (http://smpl.is.tue.mpg.de/downloads).
Run gensh.sh, and then run generate_video.py to generate a video.
sh gensh.sh
python generate_video.py
If you want to use your own data, you need provide image and the corresponding 2D joints.
You can use OpenPose or something similar to get 2D joints.
Please find more details in this code.

tf_smpl's Issues

How to use .json file?

I want to run the code in new video. I obtained the joints of the frames using openpose in .json file. How do I use .json file in place of .npz file?

when I run ./gensh.sh I got this error