This document summarizes the Faster R-CNN object detection framework. It inserts a Region Proposal Network after the last convolutional layer to directly produce region proposals rather than using external proposals. The RPN classifies anchors as object or not and regresses bounding box offsets. Proposals are then fed into Fast R-CNN for classification and further regression. Experiments show Faster R-CNN achieves real-time speeds of 0.2 seconds per image while maintaining accuracy, representing a 250x speedup over R-CNN and 25x over Fast R-CNN.

1. Speaker: Yi-Fen Li
Authors: Shaoqing Ren∗ Kaiming He Ross Girshick Jian Sun
Source: Neural Information Processing Systems (NIPS), 2015
Faster R-CNN:
Towards Real-Time Object Detection with
Region Proposal Networks
1

2. What is Convolution Neural Network (CNN)
2

3. What is Object Detection
Classification
+ Localization Object DetectionClassification
3

4. What is Object Detection
4

5. What is Object Detection
Person Detection:
p(person | box) > 0.8
5

6. What is IoU (Intersection-over-Union)
6

7. Region Proposals
7

8. Outline
• Region Proposal Networks
• Experiments
• Conclusion
8

9. Faster R-CNN
Insert a Region Proposal Network (RPN)
after the last convolutional layer1
RPN trained to produce region proposals
directly , no need for external region
Proposals.
2
After RPN , use RoI Pooling and an
upstream classifier and bbox regressor3
9

10. Faster R-CNN: Region Proposal Network
classifying regressing
True / False Box location
Slide a small window on the feature map
Build a small network for:
• classifying object or not-object
• regressing bounding box locations
Position of the sliding window provides
Localization information with reference to
the image
Box regression provides finer localization
Information with reference to this
sliding window
1
3
2
4
10

11. Faster R-CNN: Region Proposal Network
Use N anchor boxes at each location
Anchors
Anchors are translation invariant:
use the same ones at every location
Regression gives offsets from anchor
boxes
Classification gives the probability that
each(regressed) anchor shows an object
1
2
3
4
11

12. anchor boxes
12

13. Faster R-CNN: Training
In the paper: Ugly pipeline
- Use alternating optimization to train RPN,
then Fast R-CNN with RPN proposals, etc.
- More complex than it has to be
Since publication: Joint training
One network, four losses
- RPN classification (anchor good / bad)
- RPN regression (anchor -> proposal)
- Fast R-CNN classification (over classes)
- Fast R-CNN regression (proposal -> box)
13

14. Experiments
14

15. Experiments
15

16. Experiments
16

17. Experiments
17

18. Result
R-CNN Fast R-CNN Faster R-CNN
Test time per image
(with proposals)
50 seconds 2 seconds 0.2 seconds
Speedup 1x 25x 250x
mAP(voc 2007) 66.0 66.9 66.9
18

19. Thank you for listening
19