SlideShare a Scribd company logo
1 of 25
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
You Only Look Once : Unified , Real-Time Object Detection
YOLO
오혜린
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
Contents
 YOLO란  Overall
 Networkdesign
 Training
 Lossfunction
 Hyperparameter
 Test
 limitation
001 Abstract 002 Introduction 003 UnifiedDetection 004 experiments
 Objectdetection
 DPM,RCNN
 YOLO
 Comparison
 Erroranalysis
 FastRCNN+YOLO
 Generalizability
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
1
• Regression Problem
• Single Network
• Fast
• Less false positives on background
• General representations + New Domain
Abstract
YOLO란?
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
2
Introduction
Object Detection
여러 물체에 대해 어떤 물체인지, 어디에 위치해 있는지를 알아내는 작업
Classification + Localization
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
2
Introduction
Object Detection : DPM, R-CNN
문제점
1. 느리다
2. 최적화하기가 어렵다.
+ Refine
+ Eliminate
+ Rescore
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
2
Introduction
Object Detection : YOLO
사람처럼 어디에 무엇이 있는 지 한 번에 판단하자!
Single regression problem
- Single convolutional network
- Bounding box 좌표 & class 확률 계산
• 빠르다.
- Base : 45 fps , Fast : 150fps
• 전체적인 이미지를 보고 추론
- 배경을 객체로 인식하는 오류 ↓
• 일반화된 표현 학습
- 새로운 도메인에 적용 용이
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Overall
B : 각 grid마다 box 2개
Box의 중심 x, y 좌표 + box의 높이h , 너비w + confidence score
C : 각 grid마다 class 확률 계산
가장 높은 확률의 class로 정해짐
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Overall
1. 이미지를 S x S grid로 나눠 줌
- Grid cell 내 중심을 가지는 bounding boxes 존재
2. Grid cell은
B : Bounding box + Confidence Score 예측
- Bounding box : 중심의 x, y 좌표 , box의 w, h
- Confidence Score
: 객체가 존재한다면 IOU , 존재하지 않는다면 0
3. C : conditional class probabilities
- Box들과 별도로 Grid cell 각각에 대해서 class 확률 계산
- 확률이 가장 높은 class = grid cell의 class
( x, y, h, w, confidence score ) * Box의 개수
+
Class probabilities C개
S x S개의 grid cell X
본 논문에서는 S = 7 , B = 2 , C =20 로 7 x 7 x 30 tensor
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Network Design
• GoogLeNet 기반
- Inception Module❌ → 1x1 reduction layer ⭕
1x1 layers
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Network Design
Convolutional layer 24개 Fc layer 2개
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Training
• Conv layer 20개는 1000-class ImageNet 으로 pretrain
• Conv layer 4개 + Fc layer 2개 추가
- 가중치 random 초기화
• 해상도를 위해 input size 변경 224 → 448
• 마지막 layer를 제외한 모든 layer : leaky ReLU 사용
1470
reshape
Feature Extract Object Detect
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Training
𝑥𝐵1 𝑦𝐵1 ℎ𝐵1 𝑤𝐵1 𝑠𝐵1 𝑥𝐵2 𝑦𝐵2 ℎ𝐵2 𝑤𝐵2 𝑠𝐵2
Dog
0.7
cat
0.1 ………
Box1의
x,y,h,w,c-score
Box2의
x,y,h,w,c-score
해당 그리드
Class 20개의 확률
Box의 정규화 된 x, y, h, w 사용
- Box의 높이, 너비 : 전체 image에 대해 정규화
- Box의 중심좌표 x, y : grid에 대해 정규화
예시 ]
이미지의 높이 = 448, 너비 = 448
그리드의 높이 , 너비 = 64
노란색 상자의 높이 = 150
너비 = 120
중심의 좌표 x = 70 , y = 140
정규화된 box
높이 =
150
448
너비 =
120
448
x좌표 =
70 −64
64
Y좌표 =
140 −128
64
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Loss function
Bounding box coordinate regession
• 1𝑖,𝑗
𝑜𝑏𝑗
: i번째 그리드 , j번째 box가 물체가 있는 경우
• λ𝑐𝑜𝑜𝑟𝑑 : classification 보다 localization에 가중치를
주기 위한 loss term
• Box의 크기에 따른 error의 영향을 줄이기 위해
높이,너비에 대해 루트 값 사용
예시 )
29번째 그리드 내
0번째 box
Bounding box
Coordinate
regression
Confidence
score
prediction
Class
score
prediction
- SSE 사용 + 보완
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Loss function
Bounding box confidence score
prediction
• Object의 유뮤에 따라 confidence score 값 다름
• 1𝑖,𝑗
𝑜𝑏𝑗
: i번째 그리드 , j번째 box 내 물체가 있는 경우
• 1𝑖,𝑗
𝑛𝑜𝑜𝑏𝑗
: i번째 그리드 , j번째 box 내 물체가 없는 경우
• λ𝑛𝑜𝑜𝑏𝑗 = 0.5 :
- 객체가 없는 no object box가 더 많이 존재
- noobj와 obj의 균형을 맞춰 주기 위함
Bounding box
Coordinate
regression
Confidence
score
prediction
Class
score
prediction
- SSE 사용 + 보완
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Loss function
Class score prediction
• 1𝑖
𝑜𝑏𝑗
: i번째 그리드 내 물체가 있는 경우
• 물체가 없는 경우 (background로 간주)
Bounding box
Coordinate
regression
Confidence
score
prediction
Class
score
prediction
- SSE 사용 + 보완
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Hyperparameter
• epochs = 135
• batch_size = 64
• momentum = 0.9
• decay = 0.0005
• Learning rate = [0.001 , 0.01, 0.01, 0.001]
- 1epochs : 0.001 → 0.01
- 75epoch 까지 0.01 유지
- 30epoch 동안 0.001 유지
- 30epoj동안 0.0001 유지
• Avoid Overfitting → dropout , augmentation
• dropout rate = 0.5
• Data augmention
- 원본의 20% scaling , translation
- 랜덤으로 HSV 공간에서 exposure , saturation 1.5배
* HSV : 색조,채도,명도
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Test
𝑠𝐵1 𝑠𝐵2 𝑐1 ………
Box Box2 Class 20개의
확률
Class-specific confidence score 사용
- bounding box에 class가 나타날 확률과 얼마나 잘 위치해 있는지(fit) 𝑐0𝑠𝑏1
𝑐1𝑠𝑏1
𝑐2𝑠𝑏1
𝑐3𝑠𝑏1
.
.
.
.
.
𝑐18𝑠𝑏1
𝑐19𝑠𝑏1
Box별 confidence score
X
Grid 내 class의 확률
Bbox1
20x1
𝑐0𝑠𝑏2
𝑐1𝑠𝑏2
𝑐2𝑠𝑏2
𝑐3𝑠𝑏2
.
.
.
.
.
𝑐18𝑠𝑏2
𝑐19𝑠𝑏2
Bbox2
20x1
𝑐0𝑠𝑏98
𝑐1𝑠𝑏98
𝑐2𝑠𝑏98
𝑐3𝑠𝑏98
.
.
.
.
.
𝑐18𝑠𝑏98
𝑐19𝑠𝑏98
20x1
……
Bbox98
7x7개의 그리드에 대해서
box 2개씩 => 총 98개
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Test
Class prob> 0.2면,
0으로 치환 후 정렬
NMS
Class 1
Dog
bbox0 bbox1 bbox97
. . .
bbox2
0.4 0.3 0.08 0
Class 1
Dog
bbox0 bbox1 bbox97
. . .
bbox44
0.4 0.3 0.21 0 0.4 0.3 0.21 0.01 0
0 1 44 2 1
기준
IOU
기준 BOX와 선택 BOX의
IOU > 0.5 :
선택 BOX의 P = 0
20개의 class 에 대해
동일하게 반복 진행
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Test
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
3
Unified Detection
Limitation
1. 각 grid cell은 하나의 클래스만 예측
→ 물체가 겹쳐져 있다면 제대로 예측하지 못함
→ 크기가 작은 물체들은 제대로 인식되지 않음
2. Bounding box의 형태가 새롭거나 독특하면 잘 예측하지 못한다.
3. Bounding box의 크기와 상관없이 똑같이 error를 처리한다.
→ error가 작다면 large bbox는 큰 영향 x , 하지만 small bbox는 IOU에 큰 영향
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
4
Experiments
Comparision
DPM vs YOLO
- 45fps로 기존 DPM보다 빠르면서 mAP는 3배 높음
Faster R-CNN vs YOLO
- 10 mAP 낮지만 FPS가 6배 이상 높음
GoogLeNet vs VGG-16
- 같은 YOLO 모델에 convnet을 vgg16으로 교체
- 3 mAP 증가 , FPS는 2배 차이
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
4
Experiments
Error Analysis
FAST RCNN: Background Error 가 Error의 절반 가량
YOLO :Background Error가 Fast RCNN의
1
3
배
→ YOLO의 error의 대부분은 Localization으로 ,
Background error의 비율은 상당히 낮음
[ 기준 ]
Correct : class가 정확하며 IOU > 0.5 인 경우
Localization : class가 정확하고, 0.1 < IOU < 0.5 인 경우
Similar : class가 유사하고 IOU > 0.1 인 경우
Other : class는 틀렸으나, IOU > 0.1 인 경우
Background : 어떤 Object라도 IOU < 0.1 인 경우
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
4
Experiments
Fast RCNN + YOLO
FAST RCNN의 장점 : localization error ↓
+
YOLO 의 장점 : background error ↓
⇓
앙상블 사용시 mAP가 3.2% 증가
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
4
Experiments
Generalizability
- 피카소의 그림 같은 예술작품에 대해서도 person Detection이 뛰어나다.
Copyrightⓒ. Saebyeol Yu. All Rights Reserved.
• https://89douner.tistory.com/92
• https://taeu.github.io/paper/deeplearning-paper-yolo1-03/
• https://www.slideshare.net/TaegyunJeon1/pr12-you-only-look-
once-yolo-unified-realtime-object-detection

More Related Content

What's hot

Deep learning for object detection
Deep learning for object detectionDeep learning for object detection
Deep learning for object detectionWenjing Chen
 
Yolo v2 ai_tech_20190421
Yolo v2 ai_tech_20190421Yolo v2 ai_tech_20190421
Yolo v2 ai_tech_20190421穗碧 陳
 
Object Detection using Deep Neural Networks
Object Detection using Deep Neural NetworksObject Detection using Deep Neural Networks
Object Detection using Deep Neural NetworksUsman Qayyum
 
YOLO9000 - PR023
YOLO9000 - PR023YOLO9000 - PR023
YOLO9000 - PR023Jinwon Lee
 
MobileNet - PR044
MobileNet - PR044MobileNet - PR044
MobileNet - PR044Jinwon Lee
 
Modern Convolutional Neural Network techniques for image segmentation
Modern Convolutional Neural Network techniques for image segmentationModern Convolutional Neural Network techniques for image segmentation
Modern Convolutional Neural Network techniques for image segmentationGioele Ciaparrone
 
A Brief History of Object Detection / Tommi Kerola
A Brief History of Object Detection / Tommi KerolaA Brief History of Object Detection / Tommi Kerola
A Brief History of Object Detection / Tommi KerolaPreferred Networks
 
Recent Progress on Object Detection_20170331
Recent Progress on Object Detection_20170331Recent Progress on Object Detection_20170331
Recent Progress on Object Detection_20170331Jihong Kang
 
Deep learning based object detection basics
Deep learning based object detection basicsDeep learning based object detection basics
Deep learning based object detection basicsBrodmann17
 
YOLOv4: optimal speed and accuracy of object detection review
YOLOv4: optimal speed and accuracy of object detection reviewYOLOv4: optimal speed and accuracy of object detection review
YOLOv4: optimal speed and accuracy of object detection reviewLEE HOSEONG
 
AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)
AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)
AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)Fellowship at Vodafone FutureLab
 
Convolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep LearningConvolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep LearningMohamed Loey
 
Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Hwa Pyung Kim
 
Object Detection Using R-CNN Deep Learning Framework
Object Detection Using R-CNN Deep Learning FrameworkObject Detection Using R-CNN Deep Learning Framework
Object Detection Using R-CNN Deep Learning FrameworkNader Karimi
 

What's hot (20)

Deep learning for object detection
Deep learning for object detectionDeep learning for object detection
Deep learning for object detection
 
Yolo v2 ai_tech_20190421
Yolo v2 ai_tech_20190421Yolo v2 ai_tech_20190421
Yolo v2 ai_tech_20190421
 
Object Detection using Deep Neural Networks
Object Detection using Deep Neural NetworksObject Detection using Deep Neural Networks
Object Detection using Deep Neural Networks
 
Yol ov2
Yol ov2Yol ov2
Yol ov2
 
YOLO9000 - PR023
YOLO9000 - PR023YOLO9000 - PR023
YOLO9000 - PR023
 
Yolo
YoloYolo
Yolo
 
MobileNet - PR044
MobileNet - PR044MobileNet - PR044
MobileNet - PR044
 
Modern Convolutional Neural Network techniques for image segmentation
Modern Convolutional Neural Network techniques for image segmentationModern Convolutional Neural Network techniques for image segmentation
Modern Convolutional Neural Network techniques for image segmentation
 
Yolov3
Yolov3Yolov3
Yolov3
 
Yolov3
Yolov3Yolov3
Yolov3
 
A Brief History of Object Detection / Tommi Kerola
A Brief History of Object Detection / Tommi KerolaA Brief History of Object Detection / Tommi Kerola
A Brief History of Object Detection / Tommi Kerola
 
Recent Progress on Object Detection_20170331
Recent Progress on Object Detection_20170331Recent Progress on Object Detection_20170331
Recent Progress on Object Detection_20170331
 
Deep learning based object detection basics
Deep learning based object detection basicsDeep learning based object detection basics
Deep learning based object detection basics
 
YOLOv4: optimal speed and accuracy of object detection review
YOLOv4: optimal speed and accuracy of object detection reviewYOLOv4: optimal speed and accuracy of object detection review
YOLOv4: optimal speed and accuracy of object detection review
 
AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)
AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)
AlexNet(ImageNet Classification with Deep Convolutional Neural Networks)
 
Convolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep LearningConvolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep Learning
 
Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)
 
Yolo
YoloYolo
Yolo
 
Object Detection Using R-CNN Deep Learning Framework
Object Detection Using R-CNN Deep Learning FrameworkObject Detection Using R-CNN Deep Learning Framework
Object Detection Using R-CNN Deep Learning Framework
 
Yolov5
Yolov5 Yolov5
Yolov5
 

Similar to YOLO v1

IRJET - Real Time Object Detection using YOLOv3
IRJET - Real Time Object Detection using YOLOv3IRJET - Real Time Object Detection using YOLOv3
IRJET - Real Time Object Detection using YOLOv3IRJET Journal
 
auto-assistance system for visually impaired person
auto-assistance system for visually impaired personauto-assistance system for visually impaired person
auto-assistance system for visually impaired personshahsamkit73
 
Andrii Belas "Overview of object detection approaches: cases, algorithms and...
Andrii Belas  "Overview of object detection approaches: cases, algorithms and...Andrii Belas  "Overview of object detection approaches: cases, algorithms and...
Andrii Belas "Overview of object detection approaches: cases, algorithms and...Lviv Startup Club
 
#10 pydata warsaw object detection with dn ns
#10   pydata warsaw object detection with dn ns#10   pydata warsaw object detection with dn ns
#10 pydata warsaw object detection with dn nsAndrew Brozek
 
IISc Internship Report
IISc Internship ReportIISc Internship Report
IISc Internship ReportHarshilJain26
 
SEMINAR COURSE PRESENTATION on YOLO algorithm for object detection
SEMINAR COURSE PRESENTATION on YOLO algorithm for object detectionSEMINAR COURSE PRESENTATION on YOLO algorithm for object detection
SEMINAR COURSE PRESENTATION on YOLO algorithm for object detectionprasenjitroy98546
 
Review: You Only Look One-level Feature
Review: You Only Look One-level FeatureReview: You Only Look One-level Feature
Review: You Only Look One-level FeatureDongmin Choi
 
Codetecon #KRK 3 - Object detection with Deep Learning
Codetecon #KRK 3 - Object detection with Deep LearningCodetecon #KRK 3 - Object detection with Deep Learning
Codetecon #KRK 3 - Object detection with Deep LearningMatthew Opala
 
Common Understanding about YOLO
Common Understanding about YOLOCommon Understanding about YOLO
Common Understanding about YOLO재민 임
 
PR-132: SSD: Single Shot MultiBox Detector
PR-132: SSD: Single Shot MultiBox DetectorPR-132: SSD: Single Shot MultiBox Detector
PR-132: SSD: Single Shot MultiBox DetectorJinwon Lee
 
IRJET - Direct Me-Nevigation for Blind People
IRJET -  	  Direct Me-Nevigation for Blind PeopleIRJET -  	  Direct Me-Nevigation for Blind People
IRJET - Direct Me-Nevigation for Blind PeopleIRJET Journal
 
Anchor free object detection by deep learning
Anchor free object detection by deep learningAnchor free object detection by deep learning
Anchor free object detection by deep learningYu Huang
 
IRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A SurveyIRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A SurveyIRJET Journal
 
Single shot multiboxdetectors
Single shot multiboxdetectorsSingle shot multiboxdetectors
Single shot multiboxdetectors지현 백
 
20190927 generative models_aia
20190927 generative models_aia20190927 generative models_aia
20190927 generative models_aiaYi-Fan Liou
 
Visual Object Tracking: Action-Decision Networks (ADNets)
Visual Object Tracking: Action-Decision Networks (ADNets)Visual Object Tracking: Action-Decision Networks (ADNets)
Visual Object Tracking: Action-Decision Networks (ADNets)Yehya Abouelnaga
 

Similar to YOLO v1 (20)

IRJET - Real Time Object Detection using YOLOv3
IRJET - Real Time Object Detection using YOLOv3IRJET - Real Time Object Detection using YOLOv3
IRJET - Real Time Object Detection using YOLOv3
 
Yolo releases gianmaria
Yolo releases gianmariaYolo releases gianmaria
Yolo releases gianmaria
 
YOLO
YOLOYOLO
YOLO
 
auto-assistance system for visually impaired person
auto-assistance system for visually impaired personauto-assistance system for visually impaired person
auto-assistance system for visually impaired person
 
Andrii Belas "Overview of object detection approaches: cases, algorithms and...
Andrii Belas  "Overview of object detection approaches: cases, algorithms and...Andrii Belas  "Overview of object detection approaches: cases, algorithms and...
Andrii Belas "Overview of object detection approaches: cases, algorithms and...
 
#10 pydata warsaw object detection with dn ns
#10   pydata warsaw object detection with dn ns#10   pydata warsaw object detection with dn ns
#10 pydata warsaw object detection with dn ns
 
IISc Internship Report
IISc Internship ReportIISc Internship Report
IISc Internship Report
 
SEMINAR COURSE PRESENTATION on YOLO algorithm for object detection
SEMINAR COURSE PRESENTATION on YOLO algorithm for object detectionSEMINAR COURSE PRESENTATION on YOLO algorithm for object detection
SEMINAR COURSE PRESENTATION on YOLO algorithm for object detection
 
Review: You Only Look One-level Feature
Review: You Only Look One-level FeatureReview: You Only Look One-level Feature
Review: You Only Look One-level Feature
 
November 30, Projects
November 30, ProjectsNovember 30, Projects
November 30, Projects
 
Codetecon #KRK 3 - Object detection with Deep Learning
Codetecon #KRK 3 - Object detection with Deep LearningCodetecon #KRK 3 - Object detection with Deep Learning
Codetecon #KRK 3 - Object detection with Deep Learning
 
Common Understanding about YOLO
Common Understanding about YOLOCommon Understanding about YOLO
Common Understanding about YOLO
 
PR-132: SSD: Single Shot MultiBox Detector
PR-132: SSD: Single Shot MultiBox DetectorPR-132: SSD: Single Shot MultiBox Detector
PR-132: SSD: Single Shot MultiBox Detector
 
IRJET - Direct Me-Nevigation for Blind People
IRJET -  	  Direct Me-Nevigation for Blind PeopleIRJET -  	  Direct Me-Nevigation for Blind People
IRJET - Direct Me-Nevigation for Blind People
 
Anchor free object detection by deep learning
Anchor free object detection by deep learningAnchor free object detection by deep learning
Anchor free object detection by deep learning
 
IRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A SurveyIRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A Survey
 
Object Detection - Míriam Bellver - UPC Barcelona 2018
Object Detection - Míriam Bellver - UPC Barcelona 2018Object Detection - Míriam Bellver - UPC Barcelona 2018
Object Detection - Míriam Bellver - UPC Barcelona 2018
 
Single shot multiboxdetectors
Single shot multiboxdetectorsSingle shot multiboxdetectors
Single shot multiboxdetectors
 
20190927 generative models_aia
20190927 generative models_aia20190927 generative models_aia
20190927 generative models_aia
 
Visual Object Tracking: Action-Decision Networks (ADNets)
Visual Object Tracking: Action-Decision Networks (ADNets)Visual Object Tracking: Action-Decision Networks (ADNets)
Visual Object Tracking: Action-Decision Networks (ADNets)
 

Recently uploaded

FESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfFESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfMarinCaroMartnezBerg
 
100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptxAnupama Kate
 
RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998YohFuh
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingNeil Barnes
 
Industrialised data - the key to AI success.pdf
Industrialised data - the key to AI success.pdfIndustrialised data - the key to AI success.pdf
Industrialised data - the key to AI success.pdfLars Albertsson
 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130Suhani Kapoor
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPramod Kumar Srivastava
 
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Callshivangimorya083
 
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Callshivangimorya083
 
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...Sapana Sha
 
Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...
Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...
Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...shivangimorya083
 
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...Suhani Kapoor
 
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...Florian Roscheck
 
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptxEMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptxthyngster
 
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...Suhani Kapoor
 
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfMarket Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfRachmat Ramadhan H
 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...Suhani Kapoor
 

Recently uploaded (20)

FESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfFESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdf
 
100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx
 
RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998RA-11058_IRR-COMPRESS Do 198 series of 1998
RA-11058_IRR-COMPRESS Do 198 series of 1998
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data Storytelling
 
Delhi 99530 vip 56974 Genuine Escort Service Call Girls in Kishangarh
Delhi 99530 vip 56974 Genuine Escort Service Call Girls in  KishangarhDelhi 99530 vip 56974 Genuine Escort Service Call Girls in  Kishangarh
Delhi 99530 vip 56974 Genuine Escort Service Call Girls in Kishangarh
 
Industrialised data - the key to AI success.pdf
Industrialised data - the key to AI success.pdfIndustrialised data - the key to AI success.pdf
Industrialised data - the key to AI success.pdf
 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
 
Decoding Loan Approval: Predictive Modeling in Action
Decoding Loan Approval: Predictive Modeling in ActionDecoding Loan Approval: Predictive Modeling in Action
Decoding Loan Approval: Predictive Modeling in Action
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
 
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
 
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
 
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
 
Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...
Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...
Full night 🥵 Call Girls Delhi New Friends Colony {9711199171} Sanya Reddy ✌️o...
 
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
 
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...From idea to production in a day – Leveraging Azure ML and Streamlit to build...
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
 
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptxEMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
 
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
 
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfMarket Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
 
E-Commerce Order PredictionShraddha Kamble.pptx
E-Commerce Order PredictionShraddha Kamble.pptxE-Commerce Order PredictionShraddha Kamble.pptx
E-Commerce Order PredictionShraddha Kamble.pptx
 

YOLO v1

  • 1. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. You Only Look Once : Unified , Real-Time Object Detection YOLO 오혜린
  • 2. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. Contents  YOLO란  Overall  Networkdesign  Training  Lossfunction  Hyperparameter  Test  limitation 001 Abstract 002 Introduction 003 UnifiedDetection 004 experiments  Objectdetection  DPM,RCNN  YOLO  Comparison  Erroranalysis  FastRCNN+YOLO  Generalizability
  • 3. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 1 • Regression Problem • Single Network • Fast • Less false positives on background • General representations + New Domain Abstract YOLO란?
  • 4. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 2 Introduction Object Detection 여러 물체에 대해 어떤 물체인지, 어디에 위치해 있는지를 알아내는 작업 Classification + Localization
  • 5. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 2 Introduction Object Detection : DPM, R-CNN 문제점 1. 느리다 2. 최적화하기가 어렵다. + Refine + Eliminate + Rescore
  • 6. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 2 Introduction Object Detection : YOLO 사람처럼 어디에 무엇이 있는 지 한 번에 판단하자! Single regression problem - Single convolutional network - Bounding box 좌표 & class 확률 계산 • 빠르다. - Base : 45 fps , Fast : 150fps • 전체적인 이미지를 보고 추론 - 배경을 객체로 인식하는 오류 ↓ • 일반화된 표현 학습 - 새로운 도메인에 적용 용이
  • 7. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Overall B : 각 grid마다 box 2개 Box의 중심 x, y 좌표 + box의 높이h , 너비w + confidence score C : 각 grid마다 class 확률 계산 가장 높은 확률의 class로 정해짐
  • 8. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Overall 1. 이미지를 S x S grid로 나눠 줌 - Grid cell 내 중심을 가지는 bounding boxes 존재 2. Grid cell은 B : Bounding box + Confidence Score 예측 - Bounding box : 중심의 x, y 좌표 , box의 w, h - Confidence Score : 객체가 존재한다면 IOU , 존재하지 않는다면 0 3. C : conditional class probabilities - Box들과 별도로 Grid cell 각각에 대해서 class 확률 계산 - 확률이 가장 높은 class = grid cell의 class ( x, y, h, w, confidence score ) * Box의 개수 + Class probabilities C개 S x S개의 grid cell X 본 논문에서는 S = 7 , B = 2 , C =20 로 7 x 7 x 30 tensor
  • 9. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Network Design • GoogLeNet 기반 - Inception Module❌ → 1x1 reduction layer ⭕ 1x1 layers
  • 10. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Network Design Convolutional layer 24개 Fc layer 2개
  • 11. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Training • Conv layer 20개는 1000-class ImageNet 으로 pretrain • Conv layer 4개 + Fc layer 2개 추가 - 가중치 random 초기화 • 해상도를 위해 input size 변경 224 → 448 • 마지막 layer를 제외한 모든 layer : leaky ReLU 사용 1470 reshape Feature Extract Object Detect
  • 12. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Training 𝑥𝐵1 𝑦𝐵1 ℎ𝐵1 𝑤𝐵1 𝑠𝐵1 𝑥𝐵2 𝑦𝐵2 ℎ𝐵2 𝑤𝐵2 𝑠𝐵2 Dog 0.7 cat 0.1 ……… Box1의 x,y,h,w,c-score Box2의 x,y,h,w,c-score 해당 그리드 Class 20개의 확률 Box의 정규화 된 x, y, h, w 사용 - Box의 높이, 너비 : 전체 image에 대해 정규화 - Box의 중심좌표 x, y : grid에 대해 정규화 예시 ] 이미지의 높이 = 448, 너비 = 448 그리드의 높이 , 너비 = 64 노란색 상자의 높이 = 150 너비 = 120 중심의 좌표 x = 70 , y = 140 정규화된 box 높이 = 150 448 너비 = 120 448 x좌표 = 70 −64 64 Y좌표 = 140 −128 64
  • 13. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Loss function Bounding box coordinate regession • 1𝑖,𝑗 𝑜𝑏𝑗 : i번째 그리드 , j번째 box가 물체가 있는 경우 • λ𝑐𝑜𝑜𝑟𝑑 : classification 보다 localization에 가중치를 주기 위한 loss term • Box의 크기에 따른 error의 영향을 줄이기 위해 높이,너비에 대해 루트 값 사용 예시 ) 29번째 그리드 내 0번째 box Bounding box Coordinate regression Confidence score prediction Class score prediction - SSE 사용 + 보완
  • 14. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Loss function Bounding box confidence score prediction • Object의 유뮤에 따라 confidence score 값 다름 • 1𝑖,𝑗 𝑜𝑏𝑗 : i번째 그리드 , j번째 box 내 물체가 있는 경우 • 1𝑖,𝑗 𝑛𝑜𝑜𝑏𝑗 : i번째 그리드 , j번째 box 내 물체가 없는 경우 • λ𝑛𝑜𝑜𝑏𝑗 = 0.5 : - 객체가 없는 no object box가 더 많이 존재 - noobj와 obj의 균형을 맞춰 주기 위함 Bounding box Coordinate regression Confidence score prediction Class score prediction - SSE 사용 + 보완
  • 15. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Loss function Class score prediction • 1𝑖 𝑜𝑏𝑗 : i번째 그리드 내 물체가 있는 경우 • 물체가 없는 경우 (background로 간주) Bounding box Coordinate regression Confidence score prediction Class score prediction - SSE 사용 + 보완
  • 16. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Hyperparameter • epochs = 135 • batch_size = 64 • momentum = 0.9 • decay = 0.0005 • Learning rate = [0.001 , 0.01, 0.01, 0.001] - 1epochs : 0.001 → 0.01 - 75epoch 까지 0.01 유지 - 30epoch 동안 0.001 유지 - 30epoj동안 0.0001 유지 • Avoid Overfitting → dropout , augmentation • dropout rate = 0.5 • Data augmention - 원본의 20% scaling , translation - 랜덤으로 HSV 공간에서 exposure , saturation 1.5배 * HSV : 색조,채도,명도
  • 17. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Test 𝑠𝐵1 𝑠𝐵2 𝑐1 ……… Box Box2 Class 20개의 확률 Class-specific confidence score 사용 - bounding box에 class가 나타날 확률과 얼마나 잘 위치해 있는지(fit) 𝑐0𝑠𝑏1 𝑐1𝑠𝑏1 𝑐2𝑠𝑏1 𝑐3𝑠𝑏1 . . . . . 𝑐18𝑠𝑏1 𝑐19𝑠𝑏1 Box별 confidence score X Grid 내 class의 확률 Bbox1 20x1 𝑐0𝑠𝑏2 𝑐1𝑠𝑏2 𝑐2𝑠𝑏2 𝑐3𝑠𝑏2 . . . . . 𝑐18𝑠𝑏2 𝑐19𝑠𝑏2 Bbox2 20x1 𝑐0𝑠𝑏98 𝑐1𝑠𝑏98 𝑐2𝑠𝑏98 𝑐3𝑠𝑏98 . . . . . 𝑐18𝑠𝑏98 𝑐19𝑠𝑏98 20x1 …… Bbox98 7x7개의 그리드에 대해서 box 2개씩 => 총 98개
  • 18. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Test Class prob> 0.2면, 0으로 치환 후 정렬 NMS Class 1 Dog bbox0 bbox1 bbox97 . . . bbox2 0.4 0.3 0.08 0 Class 1 Dog bbox0 bbox1 bbox97 . . . bbox44 0.4 0.3 0.21 0 0.4 0.3 0.21 0.01 0 0 1 44 2 1 기준 IOU 기준 BOX와 선택 BOX의 IOU > 0.5 : 선택 BOX의 P = 0 20개의 class 에 대해 동일하게 반복 진행
  • 19. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Test
  • 20. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 3 Unified Detection Limitation 1. 각 grid cell은 하나의 클래스만 예측 → 물체가 겹쳐져 있다면 제대로 예측하지 못함 → 크기가 작은 물체들은 제대로 인식되지 않음 2. Bounding box의 형태가 새롭거나 독특하면 잘 예측하지 못한다. 3. Bounding box의 크기와 상관없이 똑같이 error를 처리한다. → error가 작다면 large bbox는 큰 영향 x , 하지만 small bbox는 IOU에 큰 영향
  • 21. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 4 Experiments Comparision DPM vs YOLO - 45fps로 기존 DPM보다 빠르면서 mAP는 3배 높음 Faster R-CNN vs YOLO - 10 mAP 낮지만 FPS가 6배 이상 높음 GoogLeNet vs VGG-16 - 같은 YOLO 모델에 convnet을 vgg16으로 교체 - 3 mAP 증가 , FPS는 2배 차이
  • 22. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 4 Experiments Error Analysis FAST RCNN: Background Error 가 Error의 절반 가량 YOLO :Background Error가 Fast RCNN의 1 3 배 → YOLO의 error의 대부분은 Localization으로 , Background error의 비율은 상당히 낮음 [ 기준 ] Correct : class가 정확하며 IOU > 0.5 인 경우 Localization : class가 정확하고, 0.1 < IOU < 0.5 인 경우 Similar : class가 유사하고 IOU > 0.1 인 경우 Other : class는 틀렸으나, IOU > 0.1 인 경우 Background : 어떤 Object라도 IOU < 0.1 인 경우
  • 23. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 4 Experiments Fast RCNN + YOLO FAST RCNN의 장점 : localization error ↓ + YOLO 의 장점 : background error ↓ ⇓ 앙상블 사용시 mAP가 3.2% 증가
  • 24. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. 4 Experiments Generalizability - 피카소의 그림 같은 예술작품에 대해서도 person Detection이 뛰어나다.
  • 25. Copyrightⓒ. Saebyeol Yu. All Rights Reserved. • https://89douner.tistory.com/92 • https://taeu.github.io/paper/deeplearning-paper-yolo1-03/ • https://www.slideshare.net/TaegyunJeon1/pr12-you-only-look- once-yolo-unified-realtime-object-detection