You are on page 1of 14

Object Detection using

Faster R-CNN
Manish Kumar
Rishon D'souza

Supervisor: Dr. Rajen Kumar Sinha


Contents
❖ Introduction
➢ What is object detection?
➢ Review of Prior Works.
❖ Faster R-CNN
➢ CNN Backbone
➢ RPN
➢ Detection Network
❖ Implementation and Applications
➢ Feature Sharing
➢ Applications
❖ Conclusion and Future Work
➢ Conclusion
➢ Phase 2 Plan
What is Object detection?
❖ Object classification vs Object localization vs Object detection
Review of Prior Works

❖ You Only Look Once (YOLO)


❖ Spatial Pyramid Pooling (SPP-Net)
❖ Single shot Detector (SSD)
❖ R-CNN family -
➢ R-CNN
➢ Fast R-CNN
➢ Faster R-CNN (RPN + Fast R-CNN)
R-CNN
The 3 main modules in R-CNN:

❖ Selective Search algorithm


❖ Each proposal is warped and then fed into the convNet, extracting a feature vector.
❖ The feature vector is passed through
➢ a trained SVM layer and
➢ a linear regressor
Fast R-CNN
Fast R-CNN improved its detection speed mainly through the following augmentations:

❖ Feature map of the entire image is computed before proposing regions, thus sharing computation.
This is done by new ROI Pooling layer.
❖ SVM is replaced with a softmax layer.
Speed Comparison of R-CNN Series
Evaluation criteria R-CNN Fast R-CNN Faster R-CNN

Test time/image 50sec 2sec 0.2sec


(with proposal)

Speedup 1x 25x 250x

mAP% 66.0 66.9 66.9


(VOC test 2007)
Conclusion

❖ We have reviewed some of the most widely used state-of-the-art object detection networks
❖ Main emphasis on the Faster R-CNN model
❖ The RPN network is able to deliver good quality proposals to the Fast R-CNN network
❖ Region proposal step is nearly cost free
Phase 2 Plan

❖ Passive Learning:
➢ As of now, Faster R-CNN model requires large amounts of labelled data for training to
achieve good results.
➢ Generally, the data labelling process is very time consuming and a costly affair.
➢ To overcome this difficulty, we will be implementing some of the Active Learning strategies in
phase 2.
Active Learning
❖ Active Learning:
➢ Iteratively selects the most informative samples having the highest impact while training the
model.
Motivation for Active Learning
❖ Unfortunately, labelling all the data points would not be possible.
❖ Randomly select a subset of data points
➢ send it for labelling
➢ train a binary classifier
➢ model obtained is sub-optimal
❖ Selection of data points for labelling was poor.
❖ Using an active learning strategy, the new decision boundary is vastly superior because it can
clearly distinguish between the two clusters.
Popular Active Learning Strategies
❖ Least Confidence: Selects the instance which has least confidence in its most likely label.
❖ Margin Sampling: Selects the instance with the least difference between the first and second most
likely labels.
❖ Entropy Sampling: The instance having the highest entropy value is sent for labelling.

Instances Label A Label B Label C

Object 1 0.35 0.4 0.25

Object 2 0.85 0.08 0.07

In the next phase, we will see some of these active learning techniques in action while training Faster
R-CNN model and explore more on the challenges existing in this area.
Thank You!
Looking forward for Phase 2 work!

You might also like