leohku / ff-emo-pipeline

Data processing pipeline for MEAD + EMOCA setup

Pipeline for processing MEAD data through EMOCA for FaceFormer training

Prerequisites

conda activate work36_cu11_pipeline

Usage

First ensure the data folder has the following directories initialised

MEAD              <- with data
MEAD_EMOCA
MEAD_FACEFORMER
MEAD_OPEN3D
MEAD_PREFORMER

Run nohup ./pipeline.sh > logs/001.log 2>&1 & to process data.

Alternatively, set your env variable BASE_DATA_PATH, then run these Python files in sequence:

cd 1-emoca
python create-templates.py
python extract-objs-from-video.py
python audio-processing.py
cd ../2-open3d
python template-normalisation.py
python mesh-registration.py
cd ../3-preformer
python template-packing.py
python sentence-packing.py