Ideas on How to Increase Speed for Obtaining Face Embeddings

First of all, thank you for this great library. I'm attempting to speed up the calculation of embeddings for faces and would be glad to hear any ideas on how to do so. Currently, I'm using the following methods:

1. A Python package:` insightface.app.FaceAnalysis`
2. 5-7 processes (until the GPU RAM is full)
3. Asynchronous loading of images
4. allowed_modules=['recognition', 'detection']

Here are some of my ideas:

1. Is it possible to calculate in mini-batches?
2. Could I use a more lightweight detector?

Start by reviewing the reported FaceAnalysis usage, including the recognition and detection modules, process count, and asynchronous image loading. Determine whether mini-batch embedding support or a lighter detector is intended, then define measurable speed and GPU-memory criteria before implementation; the issue does not name files or tests.

Written by the indexing model from the issue text.