insightface使用笔记

InsightFace 高效地实现了各种人脸识别、人脸检测和人脸对齐算法，这些算法针对训练和部署进行了优化。

 insightface 使用 onnxruntime 作为推理后端。 手动安装 onnxruntime-gpu 使用 GPU ，或者安装 onnxruntime 使用cpu

样例

import cv2
import numpy as np
import insightface
from insightface.app import FaceAnalysis
from insightface.data import get_image as ins_get_image

app = FaceAnalysis(providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
app.prepare(ctx_id=0, det_size=(640, 640))
img = ins_get_image('t1')
faces = app.get(img)
rimg = app.draw_on(img, faces)
cv2.imwrite("./t1_output.jpg", rimg)

代码将从 t1.jpg 图像中检测人脸，并在其上绘制检测结果。

模型

默认模型是buffalo_l， auto列表示是否自动下载，手动下载的，放到 ~/.insightface/models/ 下

Name	Detection Model	Recognition Model	Alignment	Attributes	Model-Size	Link	Auto
antelopev2	SCRFD-10GF	ResNet100@Glint360K	2d106 & 3d68	Gender&Age	407MB	link	N
buffalo_l	SCRFD-10GF	ResNet50@WebFace600K	2d106 & 3d68	Gender&Age	326MB	link	Y
buffalo_m	SCRFD-2.5GF	ResNet50@WebFace600K	2d106 & 3d68	Gender&Age	313MB	link	N
buffalo_s	SCRFD-500MF	MBF@WebFace600K	2d106 & 3d68	Gender&Age	159MB	link	N
buffalo_sc	SCRFD-500MF	MBF@WebFace600K	-	-	16MB	link	N

识别准确率

Name	MR-ALL	非洲	白人	南亚	东亚	LFW	CFP-FP	AgeDB-30	IJB-C(E4)
buffalo_l	91.25	90.29	94.70	93.16	74.96	99.83	99.33	98.23	97.25
buffalo_s	71.87	69.45	80.45	73.39	51.03	99.70	98.00	96.58	95.02

调用检测模型

import cv2
import numpy as np
import insightface
from insightface.app import FaceAnalysis
from insightface.data import get_image as ins_get_image

# Method-1, use FaceAnalysis
app = FaceAnalysis(allowed_modules=['detection']) # enable detection model only
app.prepare(ctx_id=0, det_size=(640, 640))

# Method-2, load model directly
detector = insightface.model_zoo.get_model('your_detection_model.onnx')
detector.prepare(ctx_id=0, input_size=(640, 640))

调用识别模型

import cv2
import numpy as np
import insightface
from insightface.app import FaceAnalysis
from insightface.data import get_image as ins_get_image

handler = insightface.model_zoo.get_model('your_recognition_model.onnx')
handler.prepare(ctx_id=0)

参考：

GitHub - deepinsight/insightface: State-of-the-art 2D and 3D Face Analysis Project

基于insightface实现的人脸识别和人脸注册_夜雨飘零1的博客-CSDN博客

 https://www.cnblogs.com/divenswu/p/16450243.html

https://github.com/deepinsight/insightface/tree/0cb03d391bb9335821448ce6e0519dbc870ec68d/examples