Notice
Recent Posts
Recent Comments
Link
«   2025/05   »
1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30 31
Archives
Today
Total
관리 메뉴

:)

Object Detection 본문

AI

Object Detection

mihee 2022. 4. 30. 19:37

Object classification : What is the object in an image

Object localization : What and where is the "single" object in an image

Object detection : What and where is the "multiple" object in an image

 

Classification

Classifier -> output : class_score [K, N]

K : The number of predicted objects

N : The number of class

 

Object localization

Regressor -> output : [K + 4]

X, Y, W, H : bounding box에 대한 정보를 담은 4개의 포인터 값

 

Sliding Window method

  • Multiple object localization
  • 이미지 안에서 작은 패치 이미지를 보고 패치를 이동해가면서 클래스와 위치 정보를 bounding box로 아웃풋을 냄.
  • Overfeat paper(패치를 여러 개 보면서 많은 연산을 가지는 게 단점)

 

- Object detection은 마지막 layer에서 fully connected layer가 아닌 2d feature map으로 나타남.

- classification의 마지막 layer는 1D 형식을 갖기 때문에 위치 정보가 없음.

 

 

stage : architecture가 한번에 forward가 진행이되는지

 

One stage Object detector architecture

 

Feed forward at once

Backbone

  •  feature extractor(이미지의 특징점들을 뽑아내면서 추상화)

 

Neck

  • Merge the different resolution feature maps

 

Dense Prediction

  • Predict score of object and bounding box
  • Regression layers

 

 

Tow stage Object detector architecture

  • 정확도는 좋지만 높은 computation cost

 

객체가 검출되는 방법

 

Grid

  • feature 맵의 pixel 수
  • predict the objects in each grid cell

Anchor

  • The detector which is predict single bounding box
  • Predict one object per anchor
  • Pre-defined bounding box shape

 

 

Object detection models

 

One stage object detectors

 

Two stage object detecors

Evaluation metric

Precision(정밀도) = tp / (tp + fp)

Recall(재현율) = tp / (tp + fn)

 

TP(True positive) :  옳은 검출

FP(Flase Positive) : 틀린 검출

FN(False Negative) : 검출되었어야 하는데 검출 안됨

TN(True Negative) : 검출되지 않아야 하는 것이 검출 안됨

 

 

F1 score로 모델 평가를 많이 함.

 

 

1에 완만하게 그려지는 그래프가 가장 좋은 모델

mAP : Mean average precision, PR curve area(적분한 값)

 

 

클래스간 간섭이 얼마나 있냐를 볼 수 있음.

'AI' 카테고리의 다른 글

Perception 기술 분류  (0) 2022.05.02
YOLOv3 시작하기  (0) 2022.05.02
CNN - Training  (0) 2022.04.29
CNN  (0) 2022.04.26
정보이론  (0) 2022.04.15
Comments