We propose a method of 3D reconstruction of a human face from video in which the 3D reconstruction algorithm and the generic model are handled separately. A 3D estimate is obtained purely from the ...

Predicting 3D human ... model. We demonstrate the benefits of our approach on four publicly available datasets (Human3.6M, HumanEva-I, AMASS, and 3DPW), outperforming the state-of-the-art.

python demo.py --load_path models/phd.ckpt-199269 --vid_id 0104 --ar_length 25 --start_frame 0 --skip_rate 5 For reference, this should be your output. To run on a generic video, you will need a ...