4
CONTENTS
1 INTRODUCTION ................................................................................................................... 6
2 YOLO – YOU ONLY LOOK ONCE ........................................................................................ 8
2.1 Concepts .................................................................................................................... 8
2.2 YOLOv1 architecture ................................................................................................ 11
2.3 Loss function ............................................................................................................ 13
3 EVOLUTIONARY HIGHLIGHTS .......................................................................................... 15
3.1 YOLOv2 ................................................................................................................... 15
3.1.1 Add Batch Normalization............................................................................ 15
3.1.2 High Resolution classifier ........................................................................... 16
3.1.3 Convolutional with anchor box ................................................................... 16
3.2 YOLOv3 ................................................................................................................... 19
3.2.1 Bigger network with ResNet ....................................................................... 19
3.2.2 Multi-scale detector .................................................................................... 21
3.3 YOLOv4 ................................................................................................................... 24
3.3.1 Object detection architecture ..................................................................... 24
3.3.2 Backbone – CSPDarknet53 ....................................................................... 26
3.3.3 Neck – Additional block – SPP block ......................................................... 28
3.3.4 Neck – Feature Aggregation – PANet ........................................................ 30
3.3.5 Head – YOLOv3 ......................................................................................... 33
3.3.6 Bag of Freebies .......................................................................................... 34
3.3.7 Bag of Specials .......................................................................................... 35
4 5
TH
GENERATION OF YOLO .............................................................................................. 36
4.1 Overview of YOLOv5 ................................................................................................ 36
4.2 Notable differences – Adaptive anchor boxes .......................................................... 37
5 IMPLEMENTING YOLOV5 ALGORITHM ............................................................................ 38
5.1 Global Wheat Head detection dataset ...................................................................... 38
5.2 Environment ............................................................................................................. 39
5.3 Preparing the dataset for training ............................................................................. 41
5.3.1 Creating the label text files ......................................................................... 44
5.3.2 Splitting data into training set and validation set ........................................ 46
