Regions with Convolutional Neural Network

lastmod 2025-02-11created 2015-12-2346 edits

R-CNN is a state-of-the-art visual object detection system that combines bottom-up region proposals with rich features computed by a convolutional neural network. At the time of its release, R-CNN improved the previous best detection performance on PASCAL VOC 2012 by 30% relative, going from 40.9% to 53.3% mean average precision. Unlike the previous best results, R-CNN achieves this performance without using contextual rescoring or an ensemble of feature types.

R-CNN

The algorithm can be summarized as follows.

The first object detection Region Proposal Network.
R-CNN generates 2,000 candidate regions (Window) and verifies each region.
Algorithms for Region Proposal include Selective Search, EdgeBox, and others.
Uses an existing CNN for Classification.
Because 2,000 overlapping regions are generated within the target image being tested, the amount of computation becomes a problem.
Each candidate region (Window) is resized to 227x227¹ and features are generated through the Network (CNN).
Finally, these values are classified with an SVM.

Its performance is good, but it has the following drawbacks.

Because the image is Warped/Cropped when the CNN classifies Region proposals, performance degrades due to image distortion/loss.
Speed degrades because about 2000 Region proposals are extracted and CNN Computation is run on all of them.
The algorithms used for Region Proposal are CPU computations that do not benefit from the GPU's fast computation. They act as a bottleneck between the CPU and GPU.

Of the three drawbacks above, the first and second were solved in SPPnet. The second drawback is solved in Fast-RCNN, introduced next, and the third drawback in Faster-RCNN.

RCNN-Object_detection_system_overview.png

Figure 1: Object detection system overview. Our system (1) takes an input image, (2) extracts around 2000 bottom-up region proposals, (3) computes features for each proposal using a large convolutional neural network (CNN), and then (4) classifies each region using class-specific linear SVMs. R-CNN achieves a mean average precision (mAP) of 53.7% on PASCAL VOC 2010. For comparison, [39] reports 35.1% mAP using the same region proposals, but with a spatial pyramid and bag-of-visual-words approach. The popular deformable part models perform at 33.4%. On the 200-class ILSVRC2013 detection dataset, R-CNN’s mAP is 31.4%, a large improvement over OverFeat [34], which had the previous best result at 24.3%.