yolov5-face 0.0.2

What's New

2024.04 ncnn-android-yolov8-face

2023.04 yolov8-face (🔥🔥🔥↑)

2022.10 yolov7-face (🔥🔥🔥↑)

2022.8: yolov7

2021.12: Yolov5-face to TensorRT.

Pytorch=1.10.0+cu102 TensorRT=8.2.0.6 Hardware=rtx2080ti

2021.11: BlazeFace

2021.08: Add new training dataset Multi-Task-Facial,improve large face detection.

QQ Group: 1164802745

Introduction

Yolov5-face is a real-time,high accuracy face detection. This fork exposes an API for face detection and facial landmark detection capabilities with minimal requirements for the API and guards for optional dependencies.

Installation

Install the package from pip, with optional dependencies cli for the cli tool and train for training:

python -m pip install yolov5-face

Face Detect CLI

To detect from a webcam or images, run:

python -m pip install yolov5-face[cli]
yolov5-face-detect --weights <model.pt> --source <0|path>

Parameters:

• --weights: model .pt path(s) (default: runs/train/exp5/weights/last.pt)
• --source: input source (default: 0 for webcam; can also be a file/folder)
• --img-size: inference size in pixels (default: 640)
• --project: output directory root (default: runs/detect)
• --name: output run name (default: exp)
• --exist-ok: reuse existing --project/--name without incrementing
• --save-img: save annotated outputs
• --view-img: display a live window

Model Architecture

Performance

Single Scale Inference on VGA resolution（max side is equal to 640 and scale).

Large family

Small family

Pretrained-Models

Data preparation

1. Download WIDERFace datasets.
2. Download annotation files from google drive.

cd data
python3 train2yolo.py /path/to/original/widerface/train [/path/to/save/widerface/train]
python3 val2yolo.py  /path/to/original/widerface [/path/to/save/widerface/val]

Training

CUDA_VISIBLE_DEVICES="0,1,2,3" python3 train.py --data data/widerface.yaml --cfg models/yolov5s.yaml --weights 'pretrained models'

WIDERFace Evaluation

python3 test_widerface.py --weights 'your test model' --img-size 640

cd widerface_evaluate
python3 evaluation.py

Test

Landmark Visulization

First row: RetinaFace, 2nd row: YOLOv5m-Face YOLO5Face was used in the 3rd place standard face recogntion track of the ICCV2021 Masked Face Recognition Challenge.

AXera demo

https://github.com/AXERA-TECH/ax-samples/blob/main/examples/ax_yolov5s_face_steps.cc

Android demo

https://github.com/FeiGeChuanShu/ncnn_Android_face/tree/main/ncnn-android-yolov5_face

OpenCV DNN demo

https://github.com/hpc203/yolov5-face-landmarks-opencv-v2

ONNXRuntime/MNN/TNN/NCNN C++ demo

https://github.com/DefTruth/lite.ai.toolkit/blob/main/lite/ort/cv/yolo5face.cpp

https://github.com/DefTruth/lite.ai.toolkit/blob/main/lite/mnn/cv/mnn_yolo5face.cpp

https://github.com/DefTruth/lite.ai.toolkit/blob/main/lite/tnn/cv/tnn_yolo5face.cpp

https://github.com/DefTruth/lite.ai.toolkit/blob/main/lite/ncnn/cv/ncnn_yolo5face.cpp

References

https://github.com/ultralytics/yolov5

https://github.com/DayBreak-u/yolo-face-with-landmark

https://github.com/xialuxi/yolov5_face_landmark

https://github.com/biubug6/Pytorch_Retinaface

https://github.com/deepinsight/insightface

Citation

• If you think this work is useful for you, please cite

  @article{YOLO5Face,
  title = {YOLO5Face: Why Reinventing a Face Detector},
  author = {Delong Qi and Weijun Tan and Qi Yao and Jingfeng Liu},
  booktitle = {ArXiv preprint ArXiv:2105.12931},
  year = {2021}
  }