This is the authors' implementation of Unsupervised Adversarial Learning of 3D Human Pose from 2D Joint Locations
- Download openpose pretrained model
- openpose_pose_coco.prototxt
- pose_iter_440000.caffemodel
- Run Inference
python bin/demo.py your/image.png --lift_model sample/gen_epoch_500.npz --model pose_iter_440000.caffemodel --proto2d openpose_pose_coco.prototxt
- Need OpenCV >= 3.4
- < 3.3 results extreamly wrong estimation
- Python 3.6.1
- Cupy 2.2.0
- Chainer 3.2.0
- ChainerUI 0.1.0
- tqdm 4.19.5
- OpenCV 3.4 with ffmpeg
- git-lfs
- to download pre-trained model
- or you can download pre-trained model directory from https://github.com/DwangoMediaVillage/3dpose_gan/blob/master/sample/gen_epoch_500.npz?raw=true
TBA
TBA