Research questions
Motivation
Previous work
Approach
Training of Faster R-CNN(4-step training)
- train RPN ( initialized with ImageNet-pre-trained model, and fine-tuned end-to-end for the region proposal task. )
- train a detector network ( initialized with ImageNet-pre-trained model ) by Fast R-CNN using the proposals generated by the Step-1 RPN
- use the detector to initialize RPN training, but fix the shared convolutional layers and only fine-tune the layers unique to RPN
- keeping the shared convolutional layers fixed, we finetune the unique layers of Fast R-CNN
Results
Summary
Contributions
- Region Proposal Networks(RPNs)
- By sharing convolutional features with the down-stream detection network, the region proposal step is nearly cost-free.
- Our method enables a unified, deep-learning-based object detection system to run at near real-time frame rates.(run at 5-17 fps).
- The learned RPN also improves region proposal quality and thus the overall object detection accuracy.
Limitations
- doesn't achieve absolute real time yet
- the accuracy should be improved further
Code
Matlab version: https://github.com/ShaoqingRen/faster_rcnn
Python version: https://github.com/rbgirshick/py-faster-rcnn