18 객체 검출 (One-stage object detection)
“부록3 매트플롯립 입문”에서 한글 폰트를 올바르게 출력하기 위한 설치 방법을 설명했다. 설치 방법은 다음과 같다.
! sudo apt - get install - y fonts - nanum * | tail - n 1 ! sudo fc - cache - fv ! rm - rf ~/ .cache / matplotlib # 필요 라이브러리 설치
! pip install torchviz | tail - n 1 ! pip install torchinfo | tail - n 1 # w = !apt install tree # print(w[-2])
모든 설치가 끝나면 한글 폰트를 바르게 출력하기 위해 [런타임] -> **[런타임 다시시작]**을 클릭한 다음, 아래 셀부터 코드를 실행해 주십시오.
# 라이브러리 임포트
% matplotlib inline import numpy as np import matplotlib.pyplot as plt from IPython.display import display
# 폰트 관련 용도 import matplotlib.font_manager as fm
# Colab, Linux # 나눔 고딕 폰트의 경로 명시 path = '/usr/share/fonts/truetype/nanum/NanumGothic.ttf' font_name = fm.FontProperties( fname = path, size = 10 ).get_name()
# Window # font_name = "NanumBarunGothic"
# Mac # font_name = "AppleGothic" # warning 표시 끄기 import warnings warnings.simplefilter( 'ignore' )
import os import numpy as np import matplotlib.pyplot as plt # 폰트 관련 용도 import matplotlib.font_manager as fm import cv2
import torch from torch import nn, optim import torchvision.transforms.functional as F from torch.utils.data import DataLoader from torchvision.io import read_image from torchvision import models, datasets, transforms from torchinfo import summary # 기본 폰트 설정 plt.rcParams[ 'font.family' ] = font_name
# 기본 폰트 사이즈 변경 plt.rcParams[ 'font.size' ] = 14
# 기본 그래프 사이즈 변경 plt.rcParams[ 'figure.figsize' ] = ( 6 , 6 )
# 기본 그리드 표시 # 필요에 따라 설정할 때는, plt.grid() plt.rcParams[ 'axes.grid' ] = True plt.rcParams[ "grid.linestyle" ] = ":"
# 마이너스 기호 정상 출력 plt.rcParams[ 'axes.unicode_minus' ] = False
# 넘파이 부동소수점 자릿수 표시 np.set_printoptions( suppress = True , precision = 4 ) # GPU 디바이스 할당
device = torch.device( "cuda:0" if torch.cuda.is_available() else "cpu" ) print (device) cuda:0
SSD 모델 사용하기
모델 불러 오기
## Pretrained object detection model list dir (models.detection) ['FCOS',
'FCOS_ResNet50_FPN_Weights',
'FasterRCNN',
'FasterRCNN_MobileNet_V3_Large_320_FPN_Weights',
'FasterRCNN_MobileNet_V3_Large_FPN_Weights',
'FasterRCNN_ResNet50_FPN_V2_Weights',
'FasterRCNN_ResNet50_FPN_Weights',
'KeypointRCNN',
'KeypointRCNN_ResNet50_FPN_Weights',
'MaskRCNN',
'MaskRCNN_ResNet50_FPN_V2_Weights',
'MaskRCNN_ResNet50_FPN_Weights',
'RetinaNet',
'RetinaNet_ResNet50_FPN_V2_Weights',
'RetinaNet_ResNet50_FPN_Weights',
'SSD300_VGG16_Weights',
'SSDLite320_MobileNet_V3_Large_Weights',
'__builtins__',
'__cached__',
'__doc__',
'__file__',
'__loader__',
'__name__',
'__package__',
'__path__',
'__spec__',
'_utils',
'anchor_utils',
'backbone_utils',
'faster_rcnn',
'fasterrcnn_mobilenet_v3_large_320_fpn',
'fasterrcnn_mobilenet_v3_large_fpn',
'fasterrcnn_resnet50_fpn',
'fasterrcnn_resnet50_fpn_v2',
'fcos',
'fcos_resnet50_fpn',
'generalized_rcnn',
'image_list',
'keypoint_rcnn',
'keypointrcnn_resnet50_fpn',
'mask_rcnn',
'maskrcnn_resnet50_fpn',
'maskrcnn_resnet50_fpn_v2',
'retinanet',
'retinanet_resnet50_fpn',
'retinanet_resnet50_fpn_v2',
'roi_heads',
'rpn',
'ssd',
'ssd300_vgg16',
'ssdlite',
'ssdlite320_mobilenet_v3_large',
'transform']
## load model # from torchvision.models.detection import ssd300_vgg16 # input image 300x300, backbone vgg16
weights = models.detection. SSD300_VGG16_Weights . COCO_V1 # 2014 ssd300 = models.detection.ssd300_vgg16( weights = weights) # weights.transforms() 모델 확인 하기
print (ssd300) SSD(
(backbone): SSDFeatureExtractorVGG(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace=True)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace=True)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU(inplace=True)
(9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace=True)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU(inplace=True)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU(inplace=True)
(16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
(17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): ReLU(inplace=True)
(19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU(inplace=True)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU(inplace=True)
)
(extra): ModuleList(
(0): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): ReLU(inplace=True)
(3): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): ReLU(inplace=True)
(5): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace=True)
(7): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(6, 6), dilation=(6, 6))
(2): ReLU(inplace=True)
(3): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1))
(4): ReLU(inplace=True)
)
)
(1): Sequential(
(0): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(3): ReLU(inplace=True)
)
(2): Sequential(
(0): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(3): ReLU(inplace=True)
)
(3-4): 2 x Sequential(
(0): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1))
(3): ReLU(inplace=True)
)
)
)
(anchor_generator): DefaultBoxGenerator(aspect_ratios=[[2], [2, 3], [2, 3], [2, 3], [2], [2]], clip=True, scales=[0.07, 0.15, 0.33, 0.51, 0.69, 0.87, 1.05], steps=[8, 16, 32, 64, 100, 300])
(head): SSDHead(
(classification_head): SSDClassificationHead(
(module_list): ModuleList(
(0): Conv2d(512, 364, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): Conv2d(1024, 546, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): Conv2d(512, 546, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): Conv2d(256, 546, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4-5): 2 x Conv2d(256, 364, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
)
(regression_head): SSDRegressionHead(
(module_list): ModuleList(
(0): Conv2d(512, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): Conv2d(1024, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): Conv2d(512, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): Conv2d(256, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4-5): 2 x Conv2d(256, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
)
)
(transform): GeneralizedRCNNTransform(
Normalize(mean=[0.48235, 0.45882, 0.40784], std=[0.00392156862745098, 0.00392156862745098, 0.00392156862745098])
Resize(min_size=(300,), max_size=300, mode='bilinear')
)
)
summary(ssd300, ( 1 , 3 , 300 , 300 )) ==========================================================================================
Layer (type:depth-idx) Output Shape Param #
==========================================================================================
SSD [97, 4] --
├─GeneralizedRCNNTransform: 1-1 [1, 3, 300, 300] --
├─SSDFeatureExtractorVGG: 1-2 [1, 256, 1, 1] 512
│ └─Sequential: 2-1 [1, 512, 38, 38] --
│ │ └─Conv2d: 3-1 [1, 64, 300, 300] (1,792)
│ │ └─ReLU: 3-2 [1, 64, 300, 300] --
│ │ └─Conv2d: 3-3 [1, 64, 300, 300] (36,928)
│ │ └─ReLU: 3-4 [1, 64, 300, 300] --
│ │ └─MaxPool2d: 3-5 [1, 64, 150, 150] --
│ │ └─Conv2d: 3-6 [1, 128, 150, 150] 73,856
│ │ └─ReLU: 3-7 [1, 128, 150, 150] --
│ │ └─Conv2d: 3-8 [1, 128, 150, 150] 147,584
│ │ └─ReLU: 3-9 [1, 128, 150, 150] --
│ │ └─MaxPool2d: 3-10 [1, 128, 75, 75] --
│ │ └─Conv2d: 3-11 [1, 256, 75, 75] 295,168
│ │ └─ReLU: 3-12 [1, 256, 75, 75] --
│ │ └─Conv2d: 3-13 [1, 256, 75, 75] 590,080
│ │ └─ReLU: 3-14 [1, 256, 75, 75] --
│ │ └─Conv2d: 3-15 [1, 256, 75, 75] 590,080
│ │ └─ReLU: 3-16 [1, 256, 75, 75] --
│ │ └─MaxPool2d: 3-17 [1, 256, 38, 38] --
│ │ └─Conv2d: 3-18 [1, 512, 38, 38] 1,180,160
│ │ └─ReLU: 3-19 [1, 512, 38, 38] --
│ │ └─Conv2d: 3-20 [1, 512, 38, 38] 2,359,808
│ │ └─ReLU: 3-21 [1, 512, 38, 38] --
│ │ └─Conv2d: 3-22 [1, 512, 38, 38] 2,359,808
│ │ └─ReLU: 3-23 [1, 512, 38, 38] --
│ └─ModuleList: 2-2 -- --
│ │ └─Sequential: 3-24 [1, 1024, 19, 19] 12,848,640
│ │ └─Sequential: 3-25 [1, 512, 10, 10] 1,442,560
│ │ └─Sequential: 3-26 [1, 256, 5, 5] 360,832
│ │ └─Sequential: 3-27 [1, 256, 3, 3] 328,064
│ │ └─Sequential: 3-28 [1, 256, 1, 1] 328,064
├─SSDHead: 1-3 [1, 8732, 91] --
│ └─SSDRegressionHead: 2-3 [1, 8732, 4] --
│ │ └─ModuleList: 3-29 -- 534,648
│ └─SSDClassificationHead: 2-4 [1, 8732, 91] --
│ │ └─ModuleList: 3-30 -- 12,163,242
├─DefaultBoxGenerator: 1-4 [8732, 4] --
==========================================================================================
Total params: 35,641,826
Trainable params: 35,603,106
Non-trainable params: 38,720
Total mult-adds (G): 34.88
==========================================================================================
Input size (MB): 1.08
Forward/backward pass size (MB): 208.89
Params size (MB): 142.57
Estimated Total Size (MB): 352.53
==========================================================================================
데이터 준비
## Image preprocessing transform = transforms.Compose([ transforms.Resize(( 300 , 300 )), transforms.ToTensor() ])
# transform = transforms.Compose([ # transforms.Resize((300, 300)), # transforms.ToTensor(), # transforms.Normalize( # mean=[0.48235, 0.45882, 0.40784], # Mean values for COCO (normalized to [0, 1]) # std=[0.00392156862745098, 0.00392156862745098, 0.00392156862745098] # Scale pixel values to [0, 1] # ) # ]) # VOCDetection 데이터셋 불러오기 dataset = datasets.VOCDetection( root = "./VOC_dataset/VOC2012" , year = "2012" , image_set = "val" , # Use "train", "val", or "trainval" download = True , transform = transform # Apply preprocessing pipeline )
# data_loader = DataLoader(dataset, batch_size=1, shuffle=False) Using downloaded and verified file: ./VOC_dataset/VOC2012\VOCtrainval_11-May-2012.tar
Extracting ./VOC_dataset/VOC2012\VOCtrainval_11-May-2012.tar to ./VOC_dataset/VOC2012
입력 영상 확인하기
idx = torch.randint( len (dataset), size = ( 1 , )) image, annotation = dataset[idx.item()]
plt.imshow(image.permute( 1 , 2 , 0 )) plt.grid( None ) plt.axis( "off" ) plt.show()
모델 예측
import time
idx = torch.randint( len (dataset), size = ( 1 , )) image_jpg, annotation = dataset[idx.item()] image = image_jpg.unsqueeze( 0 ).to(device)
ssd300.eval() start = time.time() with torch.no_grad(): prediction = ssd300(image) stop = time.time() print ( f "estimation time = { (stop - start) * 1000 :.3f } ms" ) prediction = prediction[ 0 ] # batch remove prediction estimation time = 185.503ms
{'boxes': tensor([[1.1262e+02, 6.5100e+01, 2.1513e+02, 1.8358e+02],
[8.8847e+01, 6.2713e+01, 2.4138e+02, 2.5318e+02],
[8.2444e+00, 1.4865e+02, 2.9779e+02, 2.9904e+02],
[2.4332e+02, 3.4011e+00, 2.9987e+02, 1.3884e+02],
[1.8608e+01, 2.8053e+01, 6.6836e+01, 2.1345e+02],
[6.5991e+00, 1.7761e+02, 1.5280e+02, 2.9850e+02],
[1.1509e+02, 6.3359e+01, 1.7068e+02, 1.6076e+02],
[1.5691e+02, 8.5306e+01, 2.3508e+02, 1.8724e+02],
[7.1447e+01, 7.7665e+01, 1.6359e+02, 2.0069e+02],
[2.5747e+02, 3.0218e+01, 2.9824e+02, 8.5727e+01],
[2.4039e+01, 2.4206e+02, 5.2521e+01, 2.7752e+02],
[2.7608e+01, 2.5449e+02, 4.7232e+01, 2.7788e+02],
[1.2059e+02, 1.0803e+02, 1.5563e+02, 1.8187e+02],
[1.4902e+02, 1.0802e+02, 1.9267e+02, 1.8077e+02],
[1.3566e+01, 5.9837e+01, 2.9061e+02, 2.9515e+02],
[1.3296e+02, 1.4723e+02, 2.0539e+02, 1.8141e+02],
[1.2713e+02, 1.3802e+02, 2.1133e+02, 2.0558e+02],
[3.2951e+01, 2.3434e+02, 2.3441e+02, 2.9992e+02],
[1.3268e+02, 1.2623e+02, 1.7042e+02, 1.8729e+02],
[1.1654e+02, 6.5065e+01, 1.6978e+02, 1.0808e+02],
[2.6553e+01, 4.5842e+01, 5.2569e+01, 1.5028e+02],
[1.8091e+02, 0.0000e+00, 3.0000e+02, 1.2458e+02],
[2.0823e+00, 2.0143e+01, 4.6297e+01, 2.0735e+02],
[2.4797e+02, 3.4914e+01, 2.8237e+02, 9.8553e+01],
[2.4879e+01, 1.4781e+02, 5.2215e+01, 1.9713e+02],
[1.5745e+02, 9.3629e+01, 2.0591e+02, 1.6235e+02],
[2.7077e+01, 9.3283e+01, 5.0884e+01, 1.9975e+02],
[1.1815e+02, 1.3531e+02, 1.9542e+02, 1.7165e+02],
[2.3157e+02, 1.2944e+02, 2.6351e+02, 1.9811e+02],
[2.0102e+01, 2.4127e+02, 6.5305e+01, 2.9298e+02],
[1.7710e+02, 1.0589e+02, 2.1826e+02, 1.6936e+02],
[2.2639e+02, 1.2053e+02, 2.9165e+02, 2.0699e+02],
[1.6301e+02, 1.2347e+02, 2.0437e+02, 1.8570e+02],
[1.3836e+02, 8.3384e+01, 1.9517e+02, 1.2312e+02],
[1.1054e+02, 8.4607e+01, 1.7236e+02, 1.1847e+02],
[2.0724e+02, 6.5498e+01, 2.8422e+02, 2.0185e+02],
[2.6934e+00, 2.0977e+02, 3.9960e+01, 2.8771e+02],
[2.6683e+02, 1.4120e+02, 2.9494e+02, 2.2982e+02],
[1.2180e+02, 9.9138e+01, 1.7816e+02, 1.4173e+02],
[2.0152e+02, 1.0903e+02, 2.3390e+02, 1.8048e+02],
[0.0000e+00, 2.2842e+02, 1.1478e+02, 3.0000e+02],
[2.6364e+02, 0.0000e+00, 2.9973e+02, 7.6078e+01],
[2.4596e+02, 1.2570e+02, 2.7778e+02, 2.0292e+02],
[1.9314e+01, 2.5186e+02, 3.6432e+01, 2.7902e+02],
[1.4996e+02, 1.4922e+02, 1.8758e+02, 2.0706e+02],
[1.2033e+01, 6.2718e+01, 2.9159e+02, 2.9673e+02],
[1.1332e+02, 1.1368e+02, 1.6910e+02, 1.5642e+02],
[1.3049e+02, 1.6908e+02, 2.0806e+02, 1.9631e+02],
[9.2237e+01, 4.6891e+01, 2.1070e+02, 1.2619e+02],
[2.0608e+02, 1.1134e+02, 2.3711e+02, 1.4633e+02],
[1.6260e+02, 1.1726e+02, 2.2606e+02, 1.5392e+02],
[2.1294e+02, 1.0853e+02, 2.5202e+02, 1.7428e+02],
[2.3360e+02, 7.7548e+01, 2.9801e+02, 2.5035e+02],
[7.2506e+00, 4.1083e+01, 3.4351e+01, 1.3887e+02],
[9.9559e+01, 1.0104e+02, 1.5327e+02, 1.3836e+02],
[1.0502e+02, 7.4723e+01, 1.4589e+02, 1.2955e+02],
[2.0783e+02, 3.6051e+01, 2.9338e+02, 1.9183e+02],
[1.6495e+02, 1.0280e+02, 2.2831e+02, 1.3913e+02],
[1.0848e+02, 9.8211e+01, 1.3508e+02, 1.7177e+02],
[5.5352e+01, 7.2755e+01, 1.9565e+02, 1.4870e+02],
[2.4863e+01, 3.7728e+01, 5.6120e+01, 1.0224e+02],
[1.4976e+02, 7.4522e+01, 1.8808e+02, 1.4709e+02],
[1.6498e+02, 1.1065e+02, 2.5745e+02, 2.1685e+02],
[1.5889e+02, 1.3491e+02, 2.2578e+02, 1.6964e+02],
[1.9150e+02, 1.0100e+02, 2.4268e+02, 1.4255e+02],
[3.4547e+01, 2.6295e+02, 5.0502e+01, 2.8572e+02],
[2.6314e+01, 1.4489e+02, 5.1329e+01, 2.1413e+02],
[0.0000e+00, 1.3025e+02, 2.6878e+01, 1.7329e+02],
[6.5246e+01, 1.0600e+00, 1.4453e+02, 5.3497e+01],
[1.7806e+02, 7.2294e+01, 3.0000e+02, 1.5203e+02],
[2.2755e+02, 7.5259e-01, 2.9246e+02, 8.6693e+01],
[4.9127e+01, 0.0000e+00, 1.1652e+02, 8.1133e+01],
[1.6532e+02, 1.4815e+02, 2.0229e+02, 2.0515e+02],
[1.7039e+02, 8.4908e+01, 2.2574e+02, 1.2609e+02],
[4.1368e+01, 4.8671e+01, 7.0007e+01, 1.4585e+02],
[2.2277e+02, 1.0083e+02, 2.7534e+02, 1.4145e+02],
[2.6426e+02, 1.1098e+02, 2.9474e+02, 1.9126e+02],
[3.4430e+00, 2.6317e+02, 3.3318e+01, 3.0000e+02],
[2.3276e+02, 9.4717e+01, 2.6556e+02, 1.6255e+02],
[4.3765e+01, 1.6596e+02, 1.1121e+02, 2.9378e+02],
[1.3776e+02, 1.3873e+02, 1.6569e+02, 2.2108e+02],
[5.3676e+01, 1.0546e+02, 2.0609e+02, 1.8247e+02],
[2.2202e+01, 2.5698e+02, 5.1517e+01, 2.9284e+02],
[9.3237e+01, 2.8046e+02, 1.0717e+02, 3.0000e+02],
[1.2574e+02, 1.6765e+02, 2.0966e+02, 2.3383e+02],
[0.0000e+00, 7.6399e+01, 7.8534e+01, 3.0000e+02],
[3.8097e+01, 4.0502e+01, 7.4172e+01, 9.4964e+01],
[2.0467e+02, 1.3025e+02, 3.0000e+02, 2.8190e+02],
[1.7996e+02, 1.4716e+02, 2.1229e+02, 1.8205e+02],
[2.4392e+01, 2.7371e+02, 5.7802e+01, 3.0000e+02],
[1.9919e+02, 8.7053e+01, 2.9711e+02, 2.4591e+02],
[1.2975e+02, 4.3949e+01, 2.7434e+02, 1.2061e+02],
[6.1990e+00, 7.5520e+01, 3.5306e+01, 1.7568e+02],
[1.8482e+02, 0.0000e+00, 3.0000e+02, 4.4032e+01],
[5.2050e+00, 9.1122e+01, 3.6830e+01, 1.7597e+02],
[1.0655e+02, 1.3738e+02, 1.7566e+02, 2.1477e+02],
[1.1943e+02, 1.4556e+02, 1.5214e+02, 2.0964e+02],
[2.2517e+02, 8.4079e+01, 2.7898e+02, 1.2491e+02],
[2.6137e+02, 5.0318e+01, 2.9895e+02, 1.1613e+02],
[2.7024e+02, 4.6286e+01, 2.9800e+02, 8.2859e+01],
[8.6794e+01, 2.8234e+02, 9.8415e+01, 3.0000e+02],
[7.7143e+01, 2.8459e+02, 8.8786e+01, 3.0000e+02],
[1.3975e+02, 4.1382e+01, 2.9931e+02, 2.7759e+02],
[3.4274e+00, 1.8143e+02, 3.7327e+01, 2.5543e+02],
[2.1211e+02, 1.0421e+02, 2.4400e+02, 1.3496e+02],
[1.6164e+02, 3.0446e+01, 2.9376e+02, 2.8220e+02],
[9.7141e+01, 0.0000e+00, 3.0000e+02, 7.2839e+01],
[9.0797e+00, 2.0628e+02, 1.4903e+02, 3.0000e+02],
[2.8739e+01, 2.6654e+02, 4.7588e+01, 2.8149e+02],
[7.6200e+01, 1.9732e+02, 1.4245e+02, 3.0000e+02],
[1.8015e+02, 1.5688e+02, 1.9724e+02, 1.7510e+02],
[1.0035e+02, 1.3206e+02, 1.5161e+02, 1.7284e+02],
[1.2896e+02, 1.8451e+02, 2.0478e+02, 2.1493e+02],
[1.5009e+02, 1.6989e+02, 1.8568e+02, 2.3226e+02],
[2.0381e+02, 1.1824e+02, 2.6472e+02, 1.5338e+02],
[4.5271e+01, 4.8017e+01, 5.9255e+01, 7.5684e+01],
[2.4736e+02, 1.5845e+02, 2.7825e+02, 2.3631e+02],
[0.0000e+00, 2.6236e-01, 3.9252e+01, 8.3543e+01],
[1.8883e+02, 1.4010e+02, 2.0275e+02, 1.6926e+02],
[2.3475e+02, 6.9016e+01, 2.6614e+02, 1.3673e+02],
[0.0000e+00, 1.2234e+02, 2.6402e+02, 2.9557e+02],
[9.0739e+01, 3.4670e+01, 1.5800e+02, 1.0362e+02],
[7.0053e-01, 1.9191e+02, 2.5623e+01, 2.7361e+02],
[0.0000e+00, 2.6236e-01, 3.9252e+01, 8.3543e+01],
[1.8613e+02, 1.5526e+02, 2.0546e+02, 1.7521e+02],
[1.4005e+01, 0.0000e+00, 8.4360e+01, 6.2809e+01],
[2.0693e+02, 1.1127e+02, 2.8560e+02, 2.7764e+02],
[2.1205e+01, 2.5996e+02, 3.5240e+01, 2.8987e+02],
[3.4355e+01, 4.6169e+01, 5.0254e+01, 7.4532e+01],
[2.3790e+02, 2.7117e+01, 2.6618e+02, 9.5413e+01],
[2.1104e+02, 1.1435e+02, 2.3033e+02, 1.3614e+02],
[2.3719e+01, 2.6946e+02, 3.4947e+01, 2.9880e+02],
[5.4845e+01, 3.0641e+00, 8.4185e+01, 4.3762e+01],
[1.2977e+02, 2.6673e+02, 1.6623e+02, 2.9949e+02],
[7.7951e+01, 0.0000e+00, 1.4502e+02, 7.8170e+01],
[4.4531e+01, 2.5518e+02, 5.9404e+01, 2.7930e+02],
[2.7620e+01, 2.7103e+02, 4.5167e+01, 2.9764e+02],
[2.1352e+02, 8.0281e+01, 2.5375e+02, 1.3187e+02],
[2.4486e+01, 8.5836e+01, 5.1876e+01, 1.8861e+02],
[0.0000e+00, 5.5516e+01, 1.0116e+01, 9.3727e+01],
[1.3284e+00, 1.6610e+02, 5.0052e+01, 3.0000e+02],
[1.3115e+01, 2.6681e+02, 2.7784e+01, 2.9891e+02],
[2.7369e+00, 2.0024e+02, 3.1214e+01, 2.4118e+02],
[1.6958e+02, 1.3649e+02, 2.0467e+02, 1.7422e+02],
[3.4733e+01, 2.7246e+02, 5.2063e+01, 2.9659e+02],
[1.1108e+02, 1.6895e+02, 1.7180e+02, 1.9890e+02],
[2.2348e+01, 2.0286e+02, 5.5345e+01, 2.6822e+02],
[2.1609e+02, 1.2373e+02, 2.4961e+02, 1.9544e+02],
[1.8280e+02, 1.4747e+02, 1.9556e+02, 1.6378e+02],
[4.8337e-01, 2.6548e+02, 1.8968e+01, 3.0000e+02],
[8.0369e+01, 1.7715e+02, 2.9349e+02, 2.9746e+02],
[3.6577e+01, 1.0696e+02, 5.1737e+01, 1.4984e+02],
[0.0000e+00, 1.5419e+02, 2.8511e+01, 1.8973e+02],
[1.1500e+00, 8.0133e+01, 2.2933e+01, 1.6802e+02],
[1.2527e+01, 6.0236e+01, 2.8842e+02, 2.9003e+02],
[9.2530e+00, 1.9297e+01, 3.1482e+01, 7.1072e+01],
[2.6988e+02, 5.5862e+01, 2.9933e+02, 9.7382e+01],
[1.6196e+01, 2.6468e+02, 4.6781e+01, 2.9977e+02],
[1.0586e+00, 2.3826e+02, 6.0510e+01, 3.0000e+02],
[2.2275e+02, 1.1363e+02, 2.3515e+02, 1.4041e+02],
[1.4553e+01, 0.0000e+00, 1.4937e+02, 9.3181e+01],
[1.8520e+01, 2.3740e+02, 3.6157e+01, 2.6781e+02],
[1.9543e+00, 7.3124e+01, 4.0074e+01, 1.3744e+02],
[2.5311e+01, 2.7842e+02, 3.4296e+01, 3.0000e+02],
[6.5246e+01, 1.0600e+00, 1.4453e+02, 5.3497e+01],
[6.2167e+00, 7.3394e+01, 6.4603e+01, 1.7249e+02],
[4.3474e+01, 1.1921e+02, 6.7450e+01, 2.1057e+02],
[0.0000e+00, 4.3247e+01, 1.6312e+02, 2.9152e+02],
[2.9164e+01, 6.7715e+01, 5.5653e+01, 1.1601e+02],
[6.0237e+00, 6.0450e+01, 1.9032e+01, 9.1025e+01],
[2.8704e+02, 5.2506e+01, 2.9997e+02, 8.1833e+01],
[6.1451e+00, 3.0641e+01, 1.7065e+01, 7.5105e+01],
[1.0072e+02, 2.7757e+02, 1.1564e+02, 3.0000e+02],
[2.7481e+02, 4.9030e+01, 2.9209e+02, 7.9942e+01],
[2.8040e+02, 0.0000e+00, 3.0000e+02, 1.3579e+02],
[2.6483e+02, 4.1991e+01, 2.9167e+02, 8.4969e+01],
[1.4296e+02, 7.8226e+00, 2.6265e+02, 9.8022e+01],
[2.5156e+02, 3.6799e+01, 2.8027e+02, 9.1022e+01],
[1.1677e+01, 2.0524e+02, 2.7458e+01, 2.3653e+02],
[2.8534e+02, 1.1165e+02, 2.9798e+02, 1.6020e+02],
[2.3253e+02, 1.6188e+02, 2.8907e+02, 2.0202e+02],
[1.4264e+01, 2.3510e+02, 8.2463e+01, 2.9864e+02],
[2.1017e+00, 0.0000e+00, 3.4675e+01, 8.8614e+01],
[1.7790e+00, 4.1898e+01, 2.2267e+01, 1.3345e+02],
[1.1363e+02, 2.6574e+02, 1.4332e+02, 2.9985e+02],
[2.8704e+02, 5.2506e+01, 2.9997e+02, 8.1833e+01],
[2.5011e+02, 6.3502e-01, 2.8413e+02, 7.0594e+01],
[7.0394e+01, 2.7544e+02, 9.9670e+01, 3.0000e+02],
[1.6091e+02, 1.7158e+02, 2.5516e+02, 2.9511e+02],
[2.8069e+01, 2.7779e+02, 4.3785e+01, 3.0000e+02],
[9.8555e+00, 1.1961e+02, 6.6770e+01, 2.0918e+02],
[2.2301e+02, 1.1039e+02, 2.4986e+02, 1.4078e+02],
[3.9344e+00, 2.5360e+02, 4.8506e+01, 2.9934e+02],
[2.7471e+02, 6.5201e+01, 2.9317e+02, 8.5715e+01],
[2.1740e+01, 1.9924e+02, 3.3347e+01, 2.3876e+02],
[0.0000e+00, 1.8690e+02, 2.6029e+01, 2.3328e+02],
[3.4590e+01, 2.2481e+02, 7.2581e+01, 2.8204e+02],
[0.0000e+00, 2.3549e+02, 2.8934e+01, 2.9230e+02],
[1.1500e+00, 8.0133e+01, 2.2933e+01, 1.6802e+02],
[2.9411e+02, 1.1617e+02, 3.0000e+02, 1.3939e+02]], device='cuda:0'),
'scores': tensor([0.9462, 0.1989, 0.1664, 0.1032, 0.0894, 0.0836, 0.0830, 0.0819, 0.0794,
0.0723, 0.0723, 0.0720, 0.0700, 0.0675, 0.0648, 0.0641, 0.0613, 0.0606,
0.0600, 0.0597, 0.0590, 0.0560, 0.0547, 0.0541, 0.0538, 0.0532, 0.0518,
0.0517, 0.0501, 0.0501, 0.0498, 0.0497, 0.0493, 0.0491, 0.0490, 0.0475,
0.0475, 0.0471, 0.0465, 0.0463, 0.0462, 0.0459, 0.0454, 0.0445, 0.0434,
0.0433, 0.0432, 0.0425, 0.0419, 0.0419, 0.0419, 0.0416, 0.0416, 0.0409,
0.0404, 0.0398, 0.0398, 0.0397, 0.0396, 0.0395, 0.0387, 0.0385, 0.0382,
0.0369, 0.0369, 0.0366, 0.0363, 0.0354, 0.0352, 0.0350, 0.0349, 0.0345,
0.0343, 0.0341, 0.0339, 0.0331, 0.0327, 0.0324, 0.0323, 0.0320, 0.0320,
0.0318, 0.0317, 0.0317, 0.0316, 0.0315, 0.0313, 0.0308, 0.0308, 0.0307,
0.0304, 0.0302, 0.0302, 0.0301, 0.0300, 0.0297, 0.0297, 0.0297, 0.0297,
0.0297, 0.0295, 0.0294, 0.0294, 0.0293, 0.0293, 0.0293, 0.0292, 0.0291,
0.0291, 0.0289, 0.0284, 0.0282, 0.0282, 0.0281, 0.0279, 0.0277, 0.0277,
0.0277, 0.0276, 0.0273, 0.0272, 0.0272, 0.0271, 0.0271, 0.0270, 0.0270,
0.0268, 0.0267, 0.0265, 0.0265, 0.0264, 0.0262, 0.0261, 0.0259, 0.0259,
0.0258, 0.0258, 0.0257, 0.0256, 0.0255, 0.0255, 0.0254, 0.0254, 0.0252,
0.0251, 0.0251, 0.0250, 0.0249, 0.0249, 0.0247, 0.0245, 0.0243, 0.0241,
0.0240, 0.0238, 0.0237, 0.0236, 0.0235, 0.0231, 0.0229, 0.0229, 0.0229,
0.0228, 0.0228, 0.0228, 0.0227, 0.0225, 0.0224, 0.0221, 0.0220, 0.0220,
0.0220, 0.0220, 0.0219, 0.0218, 0.0217, 0.0216, 0.0216, 0.0216, 0.0214,
0.0212, 0.0211, 0.0211, 0.0211, 0.0210, 0.0210, 0.0208, 0.0207, 0.0207,
0.0206, 0.0205, 0.0205, 0.0205, 0.0203, 0.0203, 0.0203, 0.0202, 0.0202,
0.0201, 0.0201], device='cuda:0'),
'labels': tensor([16, 16, 2, 64, 44, 2, 16, 16, 16, 86, 44, 44, 16, 16, 64, 16, 16, 2,
16, 16, 44, 64, 44, 86, 47, 16, 44, 16, 86, 44, 16, 86, 16, 16, 16, 16,
44, 86, 16, 16, 2, 64, 86, 44, 16, 61, 16, 16, 16, 62, 16, 16, 86, 44,
16, 16, 64, 16, 16, 16, 44, 16, 16, 16, 16, 44, 44, 62, 31, 16, 64, 1,
16, 16, 44, 16, 86, 62, 16, 2, 16, 16, 44, 1, 16, 2, 44, 86, 52, 62,
64, 16, 44, 64, 47, 16, 16, 16, 86, 31, 1, 1, 16, 44, 62, 64, 64, 4,
44, 2, 52, 16, 16, 16, 16, 44, 86, 31, 52, 16, 4, 16, 44, 27, 52, 1,
62, 44, 44, 86, 62, 44, 1, 47, 1, 44, 44, 16, 47, 1, 2, 62, 62, 52,
44, 16, 44, 86, 52, 62, 62, 44, 62, 47, 86, 1, 62, 62, 2, 62, 1, 44,
47, 62, 27, 47, 44, 2, 47, 1, 62, 1, 1, 62, 64, 62, 64, 62, 62, 1,
86, 4, 44, 44, 47, 31, 64, 1, 62, 62, 47, 62, 44, 31, 62, 1, 44, 44,
44, 1], device='cuda:0')}
Bounding box 그리기
image = image.cpu().data[ 0 ] image = transforms.functional.to_pil_image(image) image = np.array(image)
threshold = 0.5
for box, label, score in zip (prediction[ "boxes" ], prediction[ "labels" ], prediction[ "scores" ]): if score > threshold: box = list ( map ( int , box)) print (box) cv2.rectangle(image, (box[ 0 ], box[ 1 ]), (box[ 2 ], box[ 3 ]), ( 255 , 0 , 0 ), 2 , cv2. LINE_AA )
fig, axs = plt.subplots( 1 , 2 , figsize = ( 10 , 5 ))
axs[ 0 ].imshow(image_jpg.permute( 1 , 2 , 0 )) axs[ 0 ].grid( None ) axs[ 1 ].imshow(image) axs[ 1 ].grid( None ) plt.show() [112, 65, 215, 183]
SSD + CV DNN 얼굴검출
import numpy as np import sys import cv2 import pandas as pd img = cv2.imread( './figure/king_face.png' )
if img is None : print ( 'image read failed' ) sys.exit()
## tensorflow model model = './opencv_face_detector/opencv_face_detector_uint8.pb' config = './opencv_face_detector/opencv_face_detector.pbtxt'
face_net = cv2.dnn.readNet(model, config) # face_net.getLayerNames()
if face_net.empty(): print ( 'Net open failed' ) sys.exit()
# blobFromImage(image[, scalefactor[, size[, mean[, swapRB[, crop[, ddepth]]]]]]) -> retval blob = cv2.dnn.blobFromImage(img, 1 , ( 300 , 300 ), ( 104 , 177 , 123 ), swapRB = False )
face_net.setInput(blob) out = face_net.forward()
labels = [ "img_id" , "is_face" , "confidence" , "left" , "top" , "right" , "bottom" ] out_df = pd.DataFrame(out[ 0 ][ 0 ], columns = labels) print (out_df) img_id is_face confidence left top right bottom
0 0.0 1.0 0.989431 0.825202 0.502988 0.893821 0.658601
1 0.0 1.0 0.950553 0.147596 0.511079 0.215222 0.681570
2 0.0 1.0 0.947276 0.288199 0.444965 0.359195 0.628823
3 0.0 1.0 0.920967 0.499137 0.392986 0.589955 0.570557
4 0.0 1.0 0.835135 0.642767 0.463616 0.720524 0.659535
.. ... ... ... ... ... ... ...
195 0.0 0.0 0.000000 0.000000 0.000000 0.000000 0.000000
196 0.0 0.0 0.000000 0.000000 0.000000 0.000000 0.000000
197 0.0 0.0 0.000000 0.000000 0.000000 0.000000 0.000000
198 0.0 0.0 0.000000 0.000000 0.000000 0.000000 0.000000
199 0.0 0.0 0.000000 0.000000 0.000000 0.000000 0.000000
[200 rows x 7 columns]
detect = out[ 0 , 0 , :, :] h, w = img.shape[: 2 ]
for i in range (detect.shape[ 0 ]): confidence = detect[i, 2 ] # (0, 1, confidence, x1, y1, x2, y2) if confidence > 0.15 : # out matrix에서 x1, y1, x2, y2 값이 0 ~1로 normalize 되어 있음 x1 = int (detect[i, 3 ] * w) y1 = int (detect[i, 4 ] * h) x2 = int (detect[i, 5 ] * w) y2 = int (detect[i, 6 ] * h) cv2.rectangle(img, (x1, y1), (x2, y2), ( 0 , 0 , 255 ), 2 ) text = 'Face: {} %' .format( round (confidence * 100 , 2 )) cv2.putText(img, text, (x1, y1 - 1 ), cv2. FONT_HERSHEY_SIMPLEX , 0.8 , ( 0 , 0 , 255 ), 1 , cv2. LINE_AA ) img = cv2.cvtColor(img, cv2. COLOR_BGR2RGB )
plt.figure( figsize = ( 8 , 8 )) plt.imshow(img) plt.grid( None ) plt.axis( "off" ) plt.show()
Yolo v3 객체검출 with opencv dnn
# https://pjreddie.com/darknet/yolo/
# NMSBoxes(bboxes, scores, score_threshold, nms_threshold) -> indices # nms_threshold: nms_threshold a threshold used in non maximum suppression
# getPerfProfile() -> retval, timings # . @brief Returns overall time for inference and timings (in ticks) for layers.
# https://github.com/pjreddie/darknet/blob/master/data/coco.names ## Automating with K-Means in Python
import numpy as np from sklearn.cluster import KMeans
# Example bounding boxes (width, height) bboxes = np.array([[ 32 , 32 ], [ 64 , 64 ], [ 128 , 128 ], [ 256 , 256 ], [ 512 , 512 ], [ 128 , 64 ], [ 64 , 128 ], [ 256 , 128 ]])
# Perform K-means clustering with k=9 kmeans = KMeans( n_clusters = 3 , random_state = 0 ) kmeans.fit(bboxes)
# Output anchor boxes anchors = kmeans.cluster_centers_ print ( "Anchors:" , anchors) Anchors: [[512. 512. ]
[ 83.2 83.2]
[256. 192. ]]
import sys import numpy as np import cv2
# 모델 & 설정 파일 model = './yolo_v3_pb/yolov3.weights' config = './yolo_v3_pb/yolov3.cfg' class_labels = './yolo_v3_pb/coco.names'
# 테스트 이미지 파일 img_files = [ './figure/dog.jpg' , './figure/person.jpg' , './figure/sheep.jpg' , './figure/kite.jpg' ]
# 네트워크 생성 net = cv2.dnn.readNet(model, config)
if net.empty(): print ( 'Net open failed!' ) sys.exit()
# 클래스 이름 불러오기 classes = [] with open (class_labels, 'rt' ) as f: classes = f.read().rstrip( ' \n ' ).split( ' \n ' )
# colors = np.random.uniform(0, 255, size=(len(classes), 3)) # colors = np.array([[0, 0, 255], # [255, 0, 0], # [0, 255, 0], # [0, 255, 255], # [255, 255, 0], # [255, 0, 255]])
# 출력 레이어 이름 받아오기 net.getUnconnectedOutLayers() layer_names = net.getLayerNames() output_layers = [layer_names[i - 1 ] for i in net.getUnconnectedOutLayers()] print (output_layers) ['yolo_82', 'yolo_94', 'yolo_106']
# outs는 3개의 ndarray 리스트. # output_layers = ['yolo_82', 'yolo_94', 'yolo_106'] # output_layers[0].shape = (507, 85), 13*13*3 # output_layers[1].shape = (2028, 85), 26*26*3 # output_layers[2].shape = (8112, 85), 52*52*3 import time
confThreshold = 0.5 nmsThreshold = 0.4
# 실행 print (img_files) for i in img_files: img = cv2.imread(i)
if img is None : continue
# 블롭 생성 & 추론 blob = cv2.dnn.blobFromImage(img, 1 / 255 ., ( 320 , 320 ), swapRB = True ) # blob = cv2.dnn.blobFromImage(img, 1/255., (416, 416), swapRB=True) # blob = cv2.dnn.blobFromImage(img, 1/255., (608, 608), swapRB=True)
net.setInput(blob) outs = net.forward(output_layers)
# outs[0].shape=(507, 85), 13*13*3=507 # outs[1].shape=(2028, 85), 26*26*3=2028 # outs[2].shape=(8112, 85), 52*52*3=8112
h, w = img.shape[: 2 ]
class_ids = [] confidences = [] boxes = []
for out in outs: for detection in out: # detection: 4(bounding box) + 1(objectness_score) + 80(class confidence) scores = detection[ 5 :] class_id = np.argmax(scores) confidence = scores[class_id] if confidence > confThreshold: # 바운딩 박스 중심 좌표 & 박스 크기 cx = int (detection[ 0 ] * w) cy = int (detection[ 1 ] * h) bw = int (detection[ 2 ] * w) bh = int (detection[ 3 ] * h)
# 바운딩 박스 좌상단 좌표 sx = int (cx - bw / 2 ) sy = int (cy - bh / 2 )
boxes.append([sx, sy, bw, bh]) confidences.append( float (confidence)) class_ids.append( int (class_id))
# 비최대 억제, Non Max Suppression # https://deep-learning-study.tistory.com/403 # nmsThreshold: Determines the IoU (Intersection over Union) threshold # A higher value results in more boxes being retained. indices = cv2.dnn.NMSBoxes(boxes, confidences, confThreshold, nmsThreshold)
for i in indices: # i = i[0] sx, sy, bw, bh = boxes[i] label = f ' { classes[class_ids[i]] } : { confidences[i] :.2 } ' # color = colors[class_ids[i]] color = ( 0 , 0 , 255 ) cv2.rectangle(img, (sx, sy, bw, bh), color, 2 ) cv2.putText(img, label, (sx, sy - 10 ), cv2. FONT_HERSHEY_SIMPLEX , 0.7 , color, 2 , cv2. LINE_AA )
t, _ = net.getPerfProfile() # Total number of ticks spent during the last forward() call. label = 'Inference time: %.2f ms' % (t * 1000.0 / cv2.getTickFrequency()) cv2.putText(img, label, ( 10 , 30 ), cv2. FONT_HERSHEY_SIMPLEX , 1 , ( 0 , 0 , 255 ), 1 , cv2. LINE_AA ) img = cv2.cvtColor(img, cv2. COLOR_BGR2RGB ) plt.imshow(img) plt.grid( None ) plt.axis( "off" ) plt.show() ['./figure/dog.jpg', './figure/person.jpg', './figure/sheep.jpg', './figure/kite.jpg']
Yolo v10 객체검출 with pytorch
! pip install ultralytics from ultralytics import YOLO ## coco dataset # 클래스 이름 불러오기 # classNames = [] with open (class_labels, 'rt' ) as f: classNames = f.read().rstrip( ' \n ' ).split( ' \n ' )
# classNames = ["person", "bicycle", "car", "motorbike", "aeroplane", "bus", "train", "truck", "boat", # "traffic light", "fire hydrant", "stop sign", "parking meter", "bench", "bird", "cat", # "dog", "horse", "sheep", "cow", "elephant", "bear", "zebra", "giraffe", "backpack", "umbrella", # "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard", "sports ball", "kite", "baseball bat", # "baseball glove", "skateboard", "surfboard", "tennis racket", "bottle", "wine glass", "cup", # "fork", "knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange", "broccoli", # "carrot", "hot dog", "pizza", "donut", "cake", "chair", "sofa", "pottedplant", "bed", # "diningtable", "toilet", "tvmonitor", "laptop", "mouse", "remote", "keyboard", "cell phone", # "microwave", "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors", # "teddy bear", "hair drier", "toothbrush" # ] import time
data_dir = "./figure"
img_path = os.path.join(data_dir, "peoples.jpg" ) img = cv2.imread(img_path)
if img is None : print ( "Image read failed" ) sys.exit()
model = YOLO( "yolo11x.pt" ) # load a pretrained model (recommended for training)
start = time.time() detection = model(img, verbose = False )[ 0 ] stop = time.time() on_time = (stop - start) * 1000 print ( f "estimation time = { on_time :.3f } ms" ) fps = f ' {1000 / on_time :.4f } fps' CONFIDENCE_THRESHOLD = 0.6
for data in detection.boxes.data.tolist(): confidence = data[ 4 ] if confidence < CONFIDENCE_THRESHOLD : continue xmin, ymin, xmax, ymax = int (data[ 0 ]), int (data[ 1 ]), int (data[ 2 ]), int (data[ 3 ]) label = int (data[ 5 ]) cv2.rectangle(img, (xmin, ymin), (xmax, ymax), ( 0 , 255 , 0 ), 2 ) cv2.putText(img, classNames[label] + ' ' + str ( round (confidence, 2 )) + '%' , (xmin, ymin - 5 ), cv2. FONT_ITALIC , 0.7 , ( 0 , 0 , 255 ), 1 )
cv2.putText(img, fps, ( 10 , 20 ), cv2. FONT_HERSHEY_SIMPLEX , 0.7 , ( 0 , 0 , 255 ), 1 )
img = cv2.cvtColor(img, cv2. COLOR_BGR2RGB ) plt.imshow(img) plt.grid( None ) plt.axis( "off" ) plt.show() estimation time = 701.129ms
