This is a simple, light-weight, extremely fast face detection tool created by fal and released at fal/AuraFace-v1 on huggingface.
My contribution here is a quick and dirty "training" tool, and a "test" tool.
train.py lets you create an embedding for one or more input images of a character's face, and save this embedding to a file.
test.py lets you compare a number of images in one folder, against one or more of the saved embeddings, returning a value of how close the face matches. Basically a face similarity detection score from 0 to 1.
- Run
venv_create.bator create your own virtual environment. - Install the requirements from
requirements.txt. (pip install -r requirements.txt). Skip this step if you installed the requirements via thevenv_create.batsetup file.
Run py train.py from the (venv), to create embeddings.
One embedding will be created per subfolder in /train-input/.
To "train" multiple characters at the same time, create one folder for each character and put the face-images of each character in their respective folder.
Run py test.py from the (venv), to compare each image in /test-input/ against the embeddings you have.
You will be asked which embedding to compare the images against.
Press ENTER or 0 to compare against all, or choose the embedding with numbers as instructed.
