FaceNet
Licensing scope: See the license and provenance notice before relying on this card's license metadata for pretrained-weight redistribution or commercial use. Existing valid grants are preserved.
Model Description
facenet uses an Inception Residual Masking Network pretrained on VGGFace2 to classify facial identities. Facenet also exposes a 512 latent facial embedding space.
Model Details
- Model Type: Convolutional Neural Network (CNN)
- Architecture: Inception Residual masking network. Output layer classifies facial identities. Also provides a 512 dimensional representation layer
- Input Size: 112 x 112 pixels
- Framework: PyTorch
Model Sources
- Repository: GitHub Repository
- Paper: FaceNet: A Unified Embedding for Face Recognition and Clustering
Citation
If you use this model in your research or application, please cite the following paper:
F. Schroff, D. Kalenichenko, J. Philbin. FaceNet: A Unified Embedding for Face Recognition and Clustering, arXiv:1503.03832, 2015.
@inproceedings{schroff2015facenet,
title={Facenet: A unified embedding for face recognition and clustering},
author={Schroff, Florian and Kalenichenko, Dmitry and Philbin, James},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={815--823},
year={2015}
}
Acknowledgements
We thank Tim Esler and David Sandberg for sharing their implementation and pretrained checkpoints. Their code licenses and checkpoint/data provenance are addressed separately below.
Example Useage
import numpy as np
import torch
import torch.nn as nn
from feat.identity_detectors.facenet.facenet_model import InceptionResnetV1
from huggingface_hub import hf_hub_download
device = 'cpu'
identity_detector = InceptionResnetV1(
pretrained=None,
classify=False,
num_classes=None,
dropout_prob=0.6,
device=device,
)
identity_detector.logits = nn.Linear(512, 8631)
identity_model_file = hf_hub_download(repo_id='py-feat/facenet', filename="facenet_20180402_114759_vggface2.pth")
identity_detector.load_state_dict(torch.load(identity_model_file, map_location=device))
identity_detector.eval()
identity_detector.to(device)
# Test model
face_image = "path/to/your/test_image.jpg" # Replace with your extracted face image that is [224, 224]
# 512 dimensional Facial Embeddings
identity_embeddings = identity_detector.forward(extracted_faces)
License and provenance
This distribution retains its existing license: mit metadata, and the facenet-pytorch implementation is MIT. The shipped file is facenet_20180402_114759_vggface2.pth, corresponding to the VGGFace2-trained 20180402-114759 checkpoint in the upstream checkpoint table. CASIA-WebFace is a separate upstream checkpoint; this artifact must not be described as jointly trained on both.
VGGFace2 data/source-image rights and the exact pretrained-weight grant require separate consideration from the implementation license. Selecting FaceNet or reading an MIT badge does not establish unconditional commercial clearance.
This notice clarifies scope; it does not revoke existing valid grants or create a new license for third-party material. Dataset and teacher terms do not automatically relicense every trained artifact or inference output. A software license or model-card badge alone does not establish all checkpoint redistribution or commercial-use permissions.