OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks

By Pierre Sermanet, David Eigen, Michaël Mathieu, Rob Fergus, Yann LeCun. 2013

5,125 citations recorded by the source corpus.

We present an integrated framework for using Convolutional Networks for classification, localization and detection. We show how a multiscale and sliding window approach can be efficiently implemented within a ConvNet. We also introduce a novel deep learning approach to localization by learning to predict object boundaries. Bounding boxes are then accumulated rather than suppressed in order to increase detection confidence. We show that different tasks can be learned simultaneously using a single shared network. This integrated framework is the winner of the localization task of the ImageNet Large Scale Visual Recognition Challenge 2013 (ILSVRC2013) and obtained very competitive results for the detection and classifications tasks. In post-competition work, we establish a new state of the art for the detection task. Finally, we release a feature extractor from our best model called OverFeat.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep Residual Learning for Image Recognition (2015, background): “present a residual learning framework to ease the training of networks that are substantially deeper than those used previously.”
How transferable are features in deep neural networks? (2014, background): “…studies have taken advantage of this fact to obtain state-of-the-art results when transferring from higher layers (Donahue et al., 2013a; Zeiler and Fergus, 2013; Sermanet et al., 2014), collectively suggesting that these layers of neural networks do indeed compute features that are fairly general.”
Object Detection via a Multi-region and Semantic Segmentation-Aware CNN Model (2015, neutral): “Among them, most notable is the work of Sermanet et al. [24] with the Overfeat framework and the work of Gir-shick et al. [10] with the R-CNN framework.. Overfeat [24] uses two CNN models that apply in a sliding window fashion on multiple scales of an image.”
Fast R-CNN (2015, methodology): “Unfortunately, the approach of [10] restricts the back-propagation of errors through the network during training, potentially limiting accuracy.”
Object Detection Networks on Convolutional Feature Maps (2015, methodology): “This logical division naturally applies to a series of networks, such as AlexNet [5], Zeiler and Fergus’s (ZF) net [25], OverFeat [22], and VGG nets [15], which all have multiple fine-tunable fc layers.”
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (2015, methodology): “We further improve this strategy using the dense sliding window method in [28, 29].”
Convolutional neural networks at constrained time cost (2014, methodology): “The increased computational cost can be attributed to the increased width1 (numbers of filters) [21, 24, 1], depth (number of layers) [22, 23], smaller strides [21, 24, 22], and their combinations.”
Fully convolutional networks for semantic segmentation (2014, methodology): “These include advances in bounding box object detection [29, 10, 17], part and keypoint prediction [39, 24], and local correspondence [24, 8].”
Very Deep Convolutional Networks for Large-Scale Image Recognition (2014, methodology): “Importantly, we have made our two best-p rforming ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.”
ImageNet Large Scale Visual Recognition Challenge (2014, methodology): “1 8 O lg a R u ssa k o v sk y * et a l. ILSVRC 2013 Codename CLS LOC DET Insitutions Contributors and references Adobe 15.2 - - Adobe†, University of Illinois at Urbana-Champaign‡ Hailin Jin†, Zhe Lin†, Jianchao Yang†, Tom Paine‡ (Krizhevsky et al., 2012) AHoward 13.6 - - Andrew Howard Consulting An”
Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition (2014, methodology): “This method avoids repeatedly computing the convolutional features.”
Microsoft COCO: Common Objects in Context (2014, methodology): “For instance the ImageNet dataset [1], which contains an unprecedented number of images, has recently enabled breakthroughs in both object classiﬁcation and detection research [5], [6], [7].”
Rethinking ImageNet Pre-Training (2018, methodology): “The initial breakthrough of applying deep learning to object detection ( e.g ., R-CNN [9] and OverFeat [40]) were achieved by ﬁne-tuning networks that were pre-trained for ImageNet classiﬁcation.”
What makes ImageNet good for transfer learning? (2016, methodology): “Using ImageNet pre-trained CNN features, impressive results have been obtained on several image classification datasets [13, 35], object detection [16, 39], action recognition [41], human pose estimation [9], image segmentation [10], optical flow [47], image captioning [12, 22] and others [26].”