RCNN ( Region Based Convolution Neural Network ) is used for object detection and Localization (Identification of Objects in image with Boundaries). In this, Selective Search method (Edge Boxes and Other Region Proposal Methods) are used to determine the region proposal. This technique overcomes the problems faced using Sliding Window technique.
Around 2000 Region Proposals per Image are extracted using this technique and all these regions are Cropped and warped into fixed size (depending on the dimension which the used algorithm in the network for Classification expects) and then these regions are fed to the CNN network one by one and on the last layer of the network SVM is used for Classification.
But,
- RCNN is very slow as every single region requires forward pass of CNN and as a result we end up with too many inputs to localization network.
Top comments (0)