Target detection - yolo series algorithm

yolo algorithm

yolo algorithm idea

Yolo's network structure

network input

network output

The output is a tensor of 7*7*30

7x7 grid

30-dimensional vector

The pr(object) of an image center point grid is 1

Yolo model training

Construction of training samples

loss function

model training

model prediction

yoloSummary

yoloV2

Prediction is more accurate (better)

batch normalization

Fine-tune a classification model using high-resolution images

Sample Anchor Boxes

Cluster extraction anchor scale

yoloV2 selects the five sizes of the cluster as the anchor box

Prediction of bounding box position

Fine-grained feature fusion

multi-scale training

Faster

Identify objects more

yoloV3

Algorithm Introduction

multi-scale detection

FPN: Target detection after shallow features are fused with deep features

Network Model Structure

prior box

logistic regression

Input and output of yoloV3 model

yoloV4