camoverride / face_db

face_db

Simple script to generate a database of face embeddings from the FFHQ dataset.

Steps

Download the dataset.

Install all requirements.txt.

Run the database-generating script like this: python create_embedding_db.py path/to/face/images large.

NOTE: It takes about 2 hours to generate the database on a 2023 MacBook Pro.

Properties

If a face is unable to be embedded, it will have the value None in the database.

The resulting sqlite database will be named face_embeddings.db.

It will look like this:

id | embedding

And you can inspect it like this:

> sqlite3 face_embeddings.db
> .tables
> SELECT * FROM faces LIMIT 10;
> .exit

There are 52,001 entries, and only 88 of them fail to embed for one reason or another:

> SELECT COUNT(*) from faces;
52001
> SELECT COUNT(*) FROM faces WHERE embedding = "None";
88

If you're in the process of generating the database and want to check on the progress, use this query:

> SELECT id FROM faces ORDER BY id DESC LIMIT 1;

Enjoy 😉