D-DL4CV-Lec15e-Faster_R-CNN

Faster R-CNN

Introduction

Faster R-CNN increase the speed of fast R-CNN by using RPN instead of selection search to predict proposals from features

The other parts of faster R-CNN are exactly same as that of fast R-CNN

Region Proposal Network (RPN)

D-DL4CV-Lec15ea-RPN

RPN uses both classification and regression, but only for coarse filtering - quickly identifying "object might be here" rather than precise detection. This avoids duplicating the final stage's detailed work, making the whole pipeline efficient through clear division of labor.

Losses in Faster R-CNN

1. RPN classification: whether anchor box is object / not an object
2. RPN regression: predict transform from anchor box to proposal box
3. Object classification: classify proposals as background / object
4. Object regression: predict transform from proposal box to object box