It looks VisemeNet https://github.com/yzhou359/VisemeNet_tensorflow only runs in CUDA 8.0 environment, so I made a Docker file to run VisemeNet in CUDA 8.0 environment.
- Host: nvidia driver installed + CUDA 8 or 9(10 may work?)
- nvidia-docker2 https://github.com/NVIDIA/nvidia-docker
We've confirmed Ubutu 16.06 + CUDA 9.2 works well for the host.
Run build.sh
Clone VisemeNet somewhere and also download & store checkpoint files as described in VisemeNet's README.
$ https://github.com/yzhou359/VisemeNet_tensorflow.git
# Download checkpoint files and put it to VisemeNet_tensorflow/data/ckpt/pretrain_biwi/
Setup audio.
Run docker in CUDA 8.0 environment. Bind host's VisemeNet directory to docker container.
Edit VisemeNet path run.sh, then
$ ./run.sh
# cd /VisemeNet_tensorflow
# python3 main_test.py
mayaparam_viseme.txt will be generated in host's volume.
MIT license.