OverFeat识别定位检测-OverFeat.PDF

OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks 4 1 Pierre Sermanet David Eigen 0 Xiang Zhang Michael Mathieu Rob Fergus Yann LeCun 2 Courant Institute of Mathematical Sciences, New York University b 719 Broadway, 12th Floor, New York, NY 10003 e sermanet,deigen,xiang,mathieu,fergus,yann@ F 4 2 Abstract ] V C We present an integrated framework for using Convolutional Networks for classi- . fication, localization and detection. We show how a multiscale and sliding window s approach can be efficiently implemented within a ConvNet. We also introduce a c novel deep learning approach to localization by learning to predict object bound- [ aries. Bounding boxes are then accumulated rather than suppressed in order to 4 increase detection confidence. We show that different tasks can be learned simul- v taneously using a single shared network. This integrated framework is the winner 9 of the localization task of the ImageNet Large Scale Visual Recognition Challenge 2 2013 (ILSVRC2013) and obtained very competitive results for the detection and 2 classifications tasks. In post-competition work, we establish a new state of the art 6. for the detection task. Finally, we release a feature extractor from our best model 2