NoteNextra-origin/pages/CSE559A/CSE559A_L15.md

# CSE559A Lecture 15

## Continue on object detection

### Two strategies for object detection

#### R-CNN: Region proposals + CNN features

![R-CNN](https://notenextra.trance-0.com/CSE559A/R-CNN.png)

#### Fast R-CNN: CNN features + RoI pooling

![Fast R-CNN](https://notenextra.trance-0.com/CSE559A/Fast-R-CNN.png)

Use bilinear interpolation to get the features of the proposal.

#### Region of interest pooling

![RoI pooling](https://notenextra.trance-0.com/CSE559A/RoI-pooling.png)

Use backpropagation to get the gradient of the proposal.

### New materials

#### Faster R-CNN

Use one CNN to generate region proposals. And use another CNN to classify the proposals.