In the study R-CNN series, always see Spp-net figure, many have doubts in this paper found the answer.
Paper:Spatial Pyramid Pooling in deep convolutional Networks for Visual recognition
Transferred from: http://blog.csdn.net/xzzppp/article/details/51377731
Also refer to: http://zhangliliang.com/2014/09/13/paper-note-sppnet/
http://sinb.github.io/Whatyouknowabout/spp-net-1/
1. Introduction
Spatial pyramid pooling, so that any size of the feature map can be converted to a fixed-size eigenvector, which is the meaning of the spatial pyramid pooling (multi-scale feature extraction of fixed-size feature vectors),
Into the fully connected layer. The overall framework is as follows: input image, convolution layer extraction feature, spatial pyramid pooling to extract fixed size feature, fully connected layer.
The specific flowchart is as follows:
2, the specific algorithm of the general process
first, a search of 2000 candidate Windows is performed for the detected images by selective searching (selective search). This step is the same as the r-cnn.
Feature extraction phase. This step is the biggest difference with R-CNN, the same is the convolution neural network for feature extraction, but spp-net using pyramid pooling. This step is done in the following detail:
The entire image to be detected, input into CNN, a feature extraction, get feature maps, and then in feature maps to find the area of each candidate box, and then to each candidate box using pyramid space pool,
Extracts a fixed-length feature vector. And the R-CNN input is each candidate box, and then on the CNN, because spp-net only need to extract the entire image at once, the speed is much faster ah. The legend of the lake can be increased by 100 times times the speed,
Because R-CNN is the equivalent of traversing a CNN 2000 times, spp-net only needs to traverse 1 times. Finally, SVM algorithm is used to classify the feature vectors, which is the same as R-CNN.
3. Explanation of key steps
3.1 How to find the corresponding area of the candidate box in the original picture in feature maps
Spatial pyramid Pooling (SPP)-net (spatial pyramid Pooling) notes (RPM)