Skip to content

Latest commit

 

History

History
105 lines (75 loc) · 3.26 KB

Use_mozza_in_local.md

File metadata and controls

105 lines (75 loc) · 3.26 KB

Use Mozza in Local and offline using Docker

This is a tutorial to run you through the steps to use Mozza with python.

Install Docker Hub:

Mozza code is here : https://github.com/ducksouplab/mozza

Create a new folder called YOUR_FOLDER

Download the shape_predictor_68_face_landmarks.dat model file. It can be found online at http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2, or on debian distributions, it will have been provided by the libdlib-data package.

Check the "Use mozza with Docker" part of the documentation here. Make sure you can login to docker and pull mozza docker image:

docker login
docker pull ducksouplab/mozza:latest

If you are using arm architecture (e.g. an M1 mac ) do

docker pull ducksouplab/mozza:M1_latest

Create a new conda environement for mozza:

#create envrionemnt
conda create --name mozza python=3.9 ipython jupyter
#activate the environemnt
source activate mozza

Clone the python wrapper repository :

git clone https://github.com/Pablo-Arias/STIM.git

Add the python wrapper to the path of your environment (replace ABSOLUTE_PATH_TO_YOUR_FOLDER with your actual path):

conda develop ABSOLUTE_PATH_TO_YOUR_FOLDER/stim

Crete a folder to transfer data:

mkdir -p data/in data/out

Make sure you have the files (examples of these files can be found in the repo in "data/in".

source = "neutral.png" # A neutral face to transform
def_file = "smile10.dfm" # Deformation file to use

Create a new python script and test the following image manipulation script:

from mozza_wrapper import transform_img_with_mozza, transform_video_with_mozza

# transform one image with mozza
container_folder = "ABSOLUTE_PATH_TO_YOUR_FOLDER/data" # replace with the folder t
source = "neutral.png"
target = "smile.png"
def_file = "smile10.dfm"
transform_img_with_mozza(container_folder, source, target, wait=True, deformation_file=def_file, alpha=1, face_thresh=0.25 , overlay=False , beta=0.1, fc=5.0)

Or the following video manipulation script (add a video_in.mp4 file):

container_folder = "ABSOLUTE_PATH_TO_YOUR_FOLDER/data"
source = "video_in.mp4"
target = "video_out.mp4"
def_file = "smile10.dfm"
transform_video_with_mozza(container_folder, source, target, wait=True, deformation_file=def_file, alpha=2.5, face_thresh=0.25 , overlay=False , beta=0.1, fc=5.0)

Play with parameters. Parameters files are described here : https://github.com/ducksouplab/mozza?tab=readme-ov-file#running-the-plugin

M1 Mac users

Note : if you are running on an ARM mac, you might need to rebuild the docker image to your ARM architecture:

To do this:

#clone mozza repo
git clone https://github.com/ducksouplab/mozza.git

#move to the mozza repository
cd mozza

#build the docjer image for your platform
docker build --progress=plain -f docker/MacDockerfile.build --platform linux/arm64/v8 -t mozza:latest .

#inspect the plugin you just built
docker run -it mozza:latest gst-inspect-1.0 build/libgstmozza.so

#inspect by plugin name (since libgstaltavoce.so is in GST_PLUGIN_PATH)
docker run -it mozza:latest gst-inspect-1.0 mozza

docker tag mozza:latest ducksouplab/mozza:latest