This is the implementation of synthetic hand motion generation presented in SeqHAND (ECCV 2020).
pytorch==1.3.1
cv2==3.4.2
imageio==2.6.1
pickle
os
pillow==6.1.0
numpy
opendr>=0.76
tqdm
I have normalized BigHand dataset annotations (3D joint coordinates for each pose data). You need to get this normalized BigHand data file and save at your project directory in order to generate hand motions.
Install and set up manopth (MANO library) as explained in their work page.
Put your "mano" folder in /your/project/directory/mano/.
Hand skin colors (meshes_colored/) are obtained from our baseline work
In order to generate RGB hand motion image samples, run
python gen_sequential_synth_data.py
The genrations are saved in results/ folder.