Visage-ML

Visage-ML is a series of tools to do facial recognition on images (and create a searchable database).

Create you own database

Get a folder with images

using the following folder structure.

person_name
  - image_file_001.jpeg
  - another_01.jpeg

where person_name could be a person_name name or a UID, anything unique.

run the hasher

First we need to download some weights so the model can run.

download the following to files and put them in /hasher/weights

https://github.com/serengil/deepface_models/releases/download/v1.0/facenet512_weights.h5
https://github.com/serengil/deepface_models/releases/download/v1.0/retinaface.h5

In the hasher folder, modify the docker-compose to point to your folder of images.

docker-compose build
docker-compose up

Once complete, It creates a sidecar file called .vector next to it each image.

run the builder

To create the database that can be used to lookup, we use the builder

same thing as before, modify the docker-compose file.

docker-compose build
docker-compose up

This should give two files at the root of the folder you made available the container.

face.db and face.json these files are used by the next container.

run the face matcher

Once again. modify the docker-compose.

It creates a .json file with 10 closest results

Known issues and missing features

• No error messages are show in the UI
  • When no face can be detected
  • When no result is returned from the API
  • When creating a performer from stashbox but the local stash instance is giving a checksum error
• Scan a region of the screen to scan a single performer
• Scan for multiple performers
• On matches, link to stashbox so person can check other photo's of performer