A pipeline for creating derpfakes codecs
This is a pipeline for generating images from videos to use as input to deepfakes. To do so, it has several stages of processing:
python run.py [args]
--test
: Don't actually run any processor-intensive jobs (e.g. ffmpeg)--clearall
: Clear work and output folders before doing any work--clear
: Clear the working folder of the specified pipeline segment--verbose
: Show lots of logsThe pipeline requires source material in the two subfolders within the input
folder.
known_faces
: put images of people you want to identify here, and name each file according to the categorization you want later. e.g. Albert Einstein.jpg
and Emmy Noether.jpg
.source_videos
: and in here, put videos that contain the people you want identifiedWhen derpfakes starts, it creates a work
folder in which temporary files are stored between pipeline stages. Because each stage might be a long-running process that might crash, the files stored in the work
folder prevent work from being duplicated when derpfakes is run again.
If you want a fresh start, don't be shy to remove the work
folder. You can also run python run.py --clearall
which will destroy the work
folder and then run the pipeline. If a particular stage seems broken, you can also delete the subfolder with the name of that stage, or run python run.py --clear [stage]
. This is also a useful way to run the pipeline only after a certain stage.