py-feat/retinaface_r34

RetinaFace ResNet34

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

RetinaFace is a single-shot face detector that jointly predicts bounding boxes, 5-keypoint landmarks (eyes, nose, mouth corners), and a face/no-face score per anchor. This py-feat distribution uses a ResNet34 backbone trained on WIDERFACE, reaching 88.9% AP on WIDERFACE-Hard versus img2pose's 55.5% (per Cheong et al., Affective Science 2023). Postprocessing (priors, decode, NMS) runs fully batched on-device via torchvision.ops.batched_nms. Available in py-feat ≥ 0.7 as Detector(face_model='retinaface_r34') and MPDetector(face_model='retinaface').

Model Details

• Model Type: Convolutional Neural Network (CNN), single-shot face detector
• Architecture: ResNet34 backbone + Feature Pyramid Network + SSH context modules + 3 heads (Class / Bbox / 5-keypoint Landmark)
• Input Size: any spatial resolution (anchors generated per (H, W) and cached)
• Framework: PyTorch
• Training data: WIDERFACE (Yang et al., 2016)

Model Sources

• Repository (port used by py-feat): yakhyo/retinaface-pytorch
• Repository (original PyTorch reference): biubug6/Pytorch_Retinaface
• Paper: RetinaFace: Single-Shot Multi-Level Face Localisation in the Wild

Citation

If you use this model in your research or application, please cite the following paper:

J. Deng, J. Guo, E. Ververas, I. Kotsia, S. Zafeiriou. RetinaFace: Single-Shot Multi-Level Face Localisation in the Wild, CVPR, 2020, arXiv:1905.00641.

@inproceedings{deng2020retinaface,
  title={RetinaFace: Single-Shot Multi-Level Face Localisation in the Wild},
  author={Deng, Jiankang and Guo, Jia and Ververas, Evangelos and Kotsia, Irene and Zafeiriou, Stefanos},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={5203--5212},
  year={2020}
}

Acknowledgements

We thank Yakhyokhuja Valikhujaev for the ResNet34-backbone PyTorch implementation, biubug6 for the original PyTorch reference, and the WIDERFACE authors (Yang, Luo, Loy, Tang) for the training data.

Example Useage

import torch
from feat.face_detectors.Retinaface.Retinaface_test import Retinaface
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
from feat.utils.io import get_resource_path

device = 'cpu'
detector = Retinaface(device=device)  # py-feat helper that lazy-fetches weights from py-feat/retinaface_r34

# Or, to load weights directly:
weights_file = hf_hub_download(
    repo_id="py-feat/retinaface_r34",
    filename="model.safetensors",
    cache_dir=get_resource_path(),
)
state_dict = load_file(weights_file)
# ... pass to your own RetinaFace module instance.

License and provenance

The upstream implementation is MIT-licensed, and this distribution retains its existing license: mit metadata. The WIDER FACE dataset has separate CC BY-NC-ND terms. Those data terms must not be described as permissive merely because the detector implementation is MIT.

The software license and model-card metadata do not establish all rights in the training data or all permissions for the resulting checkpoint. Conversely, this notice does not assert that dataset terms automatically relicense trained weights or revoke existing valid grants. See the v1 component notice and dataset register for the evidence and remaining permission questions.

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.