使用 OpenCV-SeventhSense SOTA 模型進(jìn)行人臉識別
OpenCV 最近發(fā)布了與 SeventhSense 合作的人臉識別 SDK。它是 NIST 人臉識別挑戰(zhàn)賽(2022 年 3 月)的前 10 名模型,速度極快且無需 GPU。
在 opencv-seventhsense FR webapp 中,你可以創(chuàng)建一個集合并將組織中的人員聚合到組中。添加到集合中的每個人都具有姓名、出生日期、國籍、性別等屬性,可用于識別此人。
下面是 webapp 的鏈接和 UI 的快照。
OpenCV — SeventhSense-人臉識別網(wǎng)絡(luò)應(yīng)用程序
Python 和 C++ SDK 可用于將圖像對象發(fā)布到 API 并檢索給定圖像的檢測響應(yīng)。
下面的代碼片段使你能夠?qū)⒔巧砑拥郊现,并檢索給定測試圖像的檢測響應(yīng)。
我使用自己的舊照片來查看模型是否能夠準(zhǔn)確檢測到我的臉。
導(dǎo)入圖片到 FR Webapp
# Import the packages
from opencv.fr import FR
from opencv.fr.persons.schemas import PersonBase,PersonGender
from pathlib import Path
import cv2
import json
# Define the region, and developer key
BACKEND_URL = "https://us.opencv.fr"
DEVELOPER_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"
# Initialize the SDK
sdk = FR(BACKEND_URL, DEVELOPER_KEY)
directory = "/imagepath"
files = Path(directory).glob('*')
# Here we get an existing collection, but this could also
# be a collection you created
all_collections = sdk.collections.list()
all_collections = str(all_collections)
all_collections = all_collections.replace("'", """)
all_collections = json.loads(all_collections)
col_id = all_collections['collections'][0]['id']
print(f"The collection id is - {col_id}")
my_collection = sdk.collections.get(col_id)
for file in files:
if Path(str(file)).suffix == ".jpg" :
# Create PersonBase
image = cv2.imread(str(file))
person = PersonBase([image], name="Rajathithan Rajasekar",
gender=PersonGender("M"), nationality="Indian",)
# Add collection to the person's collections
person.collections.a(chǎn)ppend(my_collection)
# Create the person
person = sdk.persons.create(person)
print(f'Uploaded the image file - {file}')
從 FR webapp 人物集合中檢測給定圖像
from opencv.fr.search.schemas import DetectionRequest, SearchOptions
from opencv.fr.a(chǎn)pi_error import APIError, APIDataValidationError
from pathlib import Path
from opencv.fr import FR
import json
from json import JSONDecodeError
import cv2
import numpy as np
import imutils
import time
BACKEND_URL = "https://us.opencv.fr"
DEVELOPER_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"
# Initialize the SDK
sdk = FR(BACKEND_URL, DEVELOPER_KEY)
image_base_path = Path("sample_images")
image_path = "imagepath/test/test.jpg"
sourceimage = "imagepath/rajathithan-rajasekar.jpg"
# resize source image
simg = cv2.imread(sourceimage)
simg = imutils.resize(simg, width=640)
simg = imutils.resize(simg, height=480)
#Read old test image
frame = cv2.imread(image_path)
print(frame.shape)
#Get collection id
all_collections = sdk.collections.list()
all_collections = str(all_collections)
all_collections = all_collections.replace("'", """)
all_collections = json.loads(all_collections)
col_id = all_collections['collections'][0]['id']
print(f"The collection id is - {col_id}")
#Initilize the search options
options = SearchOptions(
collection_id=col_id,
min_score = 0.8,
)
#Detection request with search options
detect_request_with_search = DetectionRequest(image_path,options)
detectionObject = sdk.search.detect(detect_request_with_search)
print(detectionObject)
bbox = detectionObject[0].box
print(bbox)
personInfo = detectionObject[0].persons[0]
print(f"Name:{personInfo.person.name}")
print(f"Gender:{personInfo.person.gender}")
print(f"Nationality:{personInfo.person.nationality}")
print(f"Score:{personInfo.score}")
def rec_frame_display(frame: np.ndarray, roi, personInfo) -> np.ndarray:
diff1 = round((roi['bottom'] - roi['top'])/4)
diff2 = round((roi['left'] - roi['right'])/4)
cv2.line(frame, (roi['left'],roi['top']), (roi['left'],roi['top']+diff1), (0, 200, 0), 4)
cv2.line(frame, (roi['left'],roi['bottom']), (roi['left'],roi['bottom']-diff1), (0, 200, 0), 4)
cv2.line(frame, (roi['right'],roi['top']), (roi['right'],roi['top']+diff1), (0, 200, 0), 4)
cv2.line(frame, (roi['right'],roi['bottom']), (roi['right'],roi['bottom']-diff1), (0, 200, 0), 4)
cv2.line(frame, (roi['left'],roi['top']), (roi['left']-diff2,roi['top']), (0, 200, 0), 4)
cv2.line(frame, (roi['left'],roi['bottom']), (roi['left']-diff2,roi['bottom']), (0, 200, 0), 4)
cv2.line(frame, (roi['right'],roi['top']), (roi['right']+diff2,roi['top']), (0, 200, 0), 4)
cv2.line(frame, (roi['right'],roi['bottom']), (roi['right']+diff2,roi['bottom']), (0, 200, 0), 4)
cv2.putText(frame, "Name : " + personInfo.person.name, (50, 50),
cv2.FONT_HERSHEY_SIMPLEX, 2, (255, 255, 255),2,cv2.LINE_AA, False)
cv2.putText(frame, "Gender : " + str(personInfo.person.gender), (50, 150),
cv2.FONT_HERSHEY_SIMPLEX, 2, (255, 255, 255),2,cv2.LINE_AA, False)
cv2.putText(frame, "Nationality : " + str(personInfo.person.nationality), (50, 250),
cv2.FONT_HERSHEY_SIMPLEX, 2, (255, 255, 255),2,cv2.LINE_AA, False)
cv2.putText(frame, "Confidence Score : " + str(round(personInfo.score * 100,2)) + " %", (50, 350),
cv2.FONT_HERSHEY_SIMPLEX, 2, (255, 255, 255),2,cv2.LINE_AA, False)
return frame
roi = json.loads(str(bbox).replace("'","""))
frame = rec_frame_display(frame, roi, personInfo)
frame = imutils.resize(frame, width = 640,height = 480)
combine = np.concatenate((simg, frame), axis=1)
cv2.imwrite("final.jpg",combine)
cv2.imshow("Face recognition on an old pic",combine)
cv2.waitKey(0)
cv2.destroyAllWindows()
左側(cè)照片是我當(dāng)前的圖像,右側(cè)照片是我?guī)в袡z測結(jié)果的舊圖像。
原文標(biāo)題 : 使用 OpenCV-SeventhSense SOTA 模型進(jìn)行人臉識別

發(fā)表評論
最新活動更多
-
即日-9.16點(diǎn)擊進(jìn)入 >> 【限時福利】TE 2025國際物聯(lián)網(wǎng)展·深圳站
-
10月23日立即報(bào)名>> Works With 開發(fā)者大會深圳站
-
10月24日立即參評>> 【評選】維科杯·OFweek 2025(第十屆)物聯(lián)網(wǎng)行業(yè)年度評選
-
11月27日立即報(bào)名>> 【工程師系列】汽車電子技術(shù)在線大會
-
12月18日立即報(bào)名>> 【線下會議】OFweek 2025(第十屆)物聯(lián)網(wǎng)產(chǎn)業(yè)大會
-
精彩回顧立即查看>> 【限時下載】ADI中國三十周年感恩回饋助力企業(yè)升級!
推薦專題
- 1 阿里首位程序員,“掃地僧”多隆已離職
- 2 先進(jìn)算力新選擇 | 2025華為算力場景發(fā)布會暨北京xPN伙伴大會成功舉辦
- 3 宇樹機(jī)器人撞人事件的深度剖析:六維力傳感器如何成為人機(jī)安全的關(guān)鍵屏障
- 4 清華跑出具身智能獨(dú)角獸:給機(jī)器人安上眼睛和大腦,融資近20億
- 5 踢館大廠和微軟,剖析WPS靈犀的AI實(shí)用主義
- 6 特朗普要求英特爾首位華人 CEO 辭職
- 7 騰訊 Q2 財(cái)報(bào)亮眼:AI 已成第二增長曲線
- 8 AI版“四萬億刺激”計(jì)劃來了
- 9 谷歌吹響AI沖鋒號,AI還有哪些機(jī)會
- 10 蘋果把身家押在Siri上:一場輸不起的自我革命