What's New

2024.04 ncnn-android-yolov8-face

2023.04 yolov8-face (🔥🔥🔥↑)

2022.10 yolov7-face (🔥🔥🔥↑)

2022.8: yolov7

Method	Test Size	Easy	Medium	Hard	FLOPs (B) @640
yolov7-lite-t	640	88.7	85.2	71.5	0.8
yolov7-lite-s	640	92.7	89.9	78.5	3.0
yolov7-tiny	640	94.7	92.6	82.1	13.2
yolov7s	640	94.8	93.1	85.2	16.8
yolov7	640	96.9	95.5	88.0	103.4
yolov7+TTA	640	97.2	95.8	87.7	103.4

2021.12: Yolov5-face to TensorRT.

Backbone	Pytorch(ms)	TensorRT_FP16(ms)
yolov5n-0.5	7.7	2.1
yolov5n-face	7.7	2.4
yolov5s-face	5.6	2.2
yolov5m-face	9.9	3.3
yolov5l-face	15.9	4.5

Pytorch=1.10.0+cu102 TensorRT=8.2.0.6 Hardware=rtx2080ti

2021.11: BlazeFace

Method	multi scale	Easy	Medium	Hard	Model Size(MB)	Link
BlazeFace	Ture	88.5	85.5	73.1	0.472	https://github.com/PaddlePaddle/PaddleDetection
BlazeFace-FPN-SSH	Ture	90.7	88.3	79.3	0.479	https://github.com/PaddlePaddle/PaddleDetection
yolov5-blazeface	True	90.4	88.7	78.0	0.493	https://pan.baidu.com/s/1RHp8wa615OuDVhsO-qrMpQ pwd:r3v3 https://drive.google.com/file/d/1adi6ke2vCLQFcpbvFqWo_J4wZIfPqSMG
yolov5-blazeface-fpn	True	90.8	89.4	79.1	0.493	-

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

Method	Easy	Medium	Hard
YOLOv5s	94.56	92.92	83.84
YOLOv5m	95.46	93.87	85.54

Introduction

Yolov5-face is a real-time,high accuracy face detection.

Performance

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

Large family

Method	Backbone	Easy	Medium	Hard	#Params(M)	#Flops(G)
DSFD (CVPR19)	ResNet152	94.29	91.47	71.39	120.06	259.55
RetinaFace (CVPR20)	ResNet50	94.92	91.90	64.17	29.50	37.59
HAMBox (CVPR20)	ResNet50	95.27	93.76	76.75	30.24	43.28
TinaFace (Arxiv20)	ResNet50	95.61	94.25	81.43	37.98	172.95
SCRFD-34GF(Arxiv21)	Bottleneck Res	96.06	94.92	85.29	9.80	34.13
SCRFD-10GF(Arxiv21)	Basic Res	95.16	93.87	83.05	3.86	9.98
-	-	-	-	-	-	-
YOLOv5s	CSPNet	94.67	92.75	83.03	7.075	5.751
YOLOv5s6	CSPNet	95.48	93.66	82.8	12.386	6.280
YOLOv5m	CSPNet	95.30	93.76	85.28	21.063	18.146
YOLOv5m6	CSPNet	95.66	94.1	85.2	35.485	19.773
YOLOv5l	CSPNet	95.78	94.30	86.13	46.627	41.607
YOLOv5l6	CSPNet	96.38	94.90	85.88	76.674	45.279

Small family

Method	Backbone	Easy	Medium	Hard	#Params(M)	#Flops(G)
RetinaFace (CVPR20	MobileNet0.25	87.78	81.16	47.32	0.44	0.802
FaceBoxes (IJCB17)		76.17	57.17	24.18	1.01	0.275
SCRFD-0.5GF(Arxiv21)	Depth-wise Conv	90.57	88.12	68.51	0.57	0.508
SCRFD-2.5GF(Arxiv21)	Basic Res	93.78	92.16	77.87	0.67	2.53
-	-	-	-	-	-	-
YOLOv5n	ShuffleNetv2	93.74	91.54	80.32	1.726	2.111
YOLOv5n-0.5	ShuffleNetv2	90.76	88.12	73.82	0.447	0.571

Pretrained-Models

Name	Easy	Medium	Hard	FLOPs(G)	Params(M)	Link
yolov5n-0.5	90.76	88.12	73.82	0.571	0.447	Link: https://pan.baidu.com/s/1UgiKwzFq5NXI2y-Zui1kiA pwd: s5ow, https://drive.google.com/file/d/1XJ8w55Y9Po7Y5WP4X1Kg1a77ok2tL_KY/view?usp=sharing
yolov5n	93.61	91.52	80.53	2.111	1.726	Link: https://pan.baidu.com/s/1xsYns6cyB84aPDgXB7sNDQ pwd: lw9j,https://drive.google.com/file/d/18oenL6tjFkdR1f5IgpYeQfDFqU4w3jEr/view?usp=sharing
yolov5s	94.33	92.61	83.15	5.751	7.075	Link: https://pan.baidu.com/s/1fyzLxZYx7Ja1_PCIWRhxbw Link: eq0q,https://drive.google.com/file/d/1zxaHeLDyID9YU4-hqK7KNepXIwbTkRIO/view?usp=sharing
yolov5m	95.30	93.76	85.28	18.146	21.063	Link: https://pan.baidu.com/s/1oePvd2K6R4-gT0g7EERmdQ pwd: jmtk, https://drive.google.com/file/d/1Sx-KEGXSxvPMS35JhzQKeRBiqC98VDDI
yolov5l	95.78	94.30	86.13	41.607	46.627	Link: https://pan.baidu.com/s/11l4qSEgA2-c7e8lpRt8iFw pwd: 0mq7, https://drive.google.com/file/d/16F-3AjdQBn9p3nMhStUxfDNAE_1bOF_r

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}
  }