Embedded Files
Faster R-CNN
Faster R-CNN
Faster Region-based Convolutional
Neural Network
Faster R-CNN also has a large progress compared to fast R-CNN models. It uses RPN (Region Proposal Networks) instead of selective search to get proposals.
In this project, the VGG16 model will be used as a CNN model.
VGG was invented by Visual Geometry Group in Oxford. The main idea is use several 3*3 convolution kernels to replace big convolution kernels to improve the effect.
VGG16 use two 3*3 convolution kernels to replace 5*5 convolution kernels and use three 3*3 convolution kernels to replace 7*7 convolution kernels. In VGG16 model, there are 13 convolutional layer, 3 fully connected layer and 5 pool layer.
After the convolution layer, we can get (M/16)x(N/16) feature maps, and these feature maps will be used to get proposals.
Faster R-CNN totally have four different layers:
Convolution layer
Convolution layer
RPN layer
RPN layer
ROI pooling layer
ROI pooling layer
Classifier layer
Classifier layer
Page updated
Google Sites
Report abuse