Details of Detection

R-CNN

1. Input image

2. Extract Region proposals - (sliding window, selective search)

3. Compute CNN features

4. Classify Regions

 

SPPNet

1. Input image

2. Compute CNN features 

3. Spatial pyramid pooling

4. Classify Regions

 

Fast R-CNN

1. Inpute image

2. Compute CNN features

3. RoI projection (feature map의 RoI계산)

4. RoI Pooling을 통해 일정한 크기의 feature 추출

5. Classify Regions + Bounding Box Regressor(Smooth L1으로 loss계산)

 

Faster R-CNN - 현시점 끝판왕

1. Input image

2. Compute CNN features

3. RoI계산 (RPN(Region Proposal Network)이용 - anchor box 개념 사용 + selective search 대체)

4. NMS(Non Maximum Suppresion)

5. Classify Regions + Bounding Box Regressor(Smooth L1으로 loss계산)

 

SAM이 나와주길