---
library_name: pytorch
---

![retinaface_logo](resource/RetinaFace_resnet.png)

RetinaFace is a state-of-the-art single-stage face detector that combines a robust backbone with multi-task learning to predict face locations, landmarks, and context-aware features simultaneously.

Original paper: [RetinaFace: Single-Shot Multi-Level Face Localisation in the Wild](https://arxiv.org/abs/1905.00641)

# RetinaFace-Resnet50

This model uses RetinaFace with a ResNet-50 backbone, providing a strong trade-off between detection accuracy and computational efficiency. It is well suited for face detection and landmark localization in applications such as authentication, video surveillance, and human–computer interaction.

Model Configuration:
- Reference implementation: [Official RetinaFace source code](https://github.com/biubug6/Pytorch_Retinaface)
- Original Weight: [RetinaFace weight](https://drive.google.com/file/d/14KX6VqF69MdSPk3Tr9PlDYbq7ArpdNUW/view?usp=drive_link)
- Resolution: 3x640x640
- Support Cooper version:
    - Cooper SDK: [2.5.2]
    - Cooper Foundry: [2.2]

| Model | Device | Model Link |
| :-----: | :-----: | :-----: |
| RetinaFace-Resnet50 | N1-655 | [Model_Link](https://huggingface.co/Ambarella/RetinaFace/blob/main/n1-655_retinaface_resnet50.bin) |
| RetinaFace-Resnet50 | CV72 | [Model_Link](https://huggingface.co/Ambarella/RetinaFace/blob/main/cv72_retinaface_resnet50.bin) |
| RetinaFace-Resnet50 | CV75 | [Model_Link](https://huggingface.co/Ambarella/RetinaFace/blob/main/cv75_retinaface_resnet50.bin) |