README.md

Facial Feature Extractor

This repository contains a server and client to run extraction of facial templates 😀.

🖥️ Server

Note: You must be part of the sudoers to run the code below without sudo. See here for details.

To launch the backend, you need to use the server.sh script.

Some options available are shown in the table below:

Option	Description	Default	Required
--start-port	Starting port range for containers	18081	No
--gpus-start	GPU device number to start with	0	No
--gpus	Number of gpus, starting from gpus-start, to use	1	No
--workers	Number of backend processes per container	1	No
--fp16	Use half-point precision	True	No
--detection-model	Model use for face detection	scrfd_10g_gnkps	No
--recognition-model	Model use for facial feature extraction	glintr100	No
--models-path	Directory where pre-downloaded models can be found	$PWD/models	No
--data-path	Path where the data is located	None	Yes

As an example, for 1 container (1 GPU / 1 worker), with the data directory being /path/to/sample/data, you would do:

./server.sh --data-path /path/to/sample/data

The data path must be the same for the client and the server.

Note: If the container fails because it is missing the pydantic module, add its entry to the InsightFace-Server/src/requirements.txt file.

💻 Client

Note: You must be part of the sudoers to run the code below without sudo. See here for details.

Navigate to the InsightFace-Client directory first and then build the image.

cd InsightFace-Client
docker build -t xavier/arcface:v0.8.2.0 .

To start the client, you should use the docker run command. In the example below, we specify the data directory as /data and the output directory as /output. Within the output directory, you will find cropped faces (if requested with the --save-crops flag) in the /output/crops subdirectory, templates in the /output/templates subdirectory, and a summary of enrollments and failures in /output/summary. Please note that to ensure crops are saved, regardless of the flag, the mode should be set to either all or detect_only. The default mode is all.

docker run --rm \
-v /home/xavier/Documents/git/ArcFace/sample:/data \
-v /home/xavier/Documents/git/ArcFace/output:/output \
--net host \
xavier/arcface:v0.8.2.0 \
--exclude \
--save-crops \
--extension ".JPG"

Additional options can be appended to the command. Invoke the command with -h to see more information.

docker run --rm \
-v /home/xavier/Documents/git/ArcFace/sample:/data \
-v /home/xavier/Documents/git/ArcFace/output:/output \
--net host \
xavier/arcface:v0.8.2.0 -h

Alternatively, see the table below:

Option	Description	Default	Required
--host	Server hostname or IP	localhost	No
--port	Server port	18081	No
--threads	Number of worker processes for data dispatch	10	No
--batch	Number of images per server request	1	No
--dir	Path to directory with images	/data	No
--output	Directory where templates are saved	/output	No
--exclude	Exclude images with no face detected	False	No
--save-crops	Saves crops when face detected	False	No
--extension	Allowed image extensions	'.jpeg' '.jpg' '.bmp' '.png' '.webp' '.tiff'	No