MagFace

We are using the official implementation of MagFace¹ in Pytorch. See https://github.com/IrvingMeng/MagFace.

Pre-trained models

mozuma.models.magface.pretrained.torch_magface

Pre-trained MagFace module

Parameters:

Name	Type	Description	Default
device	torch.device	Torch device to initialise the model weights	required
remove_bad_faces	bool	Whether to remove the faces with bad quality from the output. This will replace features of bad faces with float("nan"). Defaults to False.	required
magnitude_threshold	float	Threshold to remove bad quality faces. The higher the stricter. Defaults to 22.5.	required

Base models

The MagFace model is an implementation of a TorchModel.

mozuma.models.magface.modules.TorchMagFaceModule

MagFace face embeddings from MTCNN detected faces

The input dataset should return a tuple of image data and bounding box information

Attributes:

Name	Type	Description
device	torch.device	Torch device to initialise the model weights
remove_bad_faces	bool	Whether to remove the faces with bad quality from the output. This will replace features of bad faces with float("nan"). Defaults to False.
magnitude_threshold	float	Threshold to remove bad quality faces. The higher the stricter. Defaults to 22.5.

Provider store

See the stores documentation for usage.

mozuma.models.magface.stores.MagFaceStore

Pre-trained model states by MagFace (https://github.com/IrvingMeng/MagFace)

These are identified by training_id=magface.

---

1. Qiang Meng, Shichao Zhao, Zhida Huang, and Feng Zhou. Magface: a universal representation for face recognition and quality assessment. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 14225–14234. June 2021. ↩