OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks

TL;DR

OverFeat introduces an integrated convolutional network framework capable of simultaneous classification, localization, and detection, achieving top results in the 2013 ImageNet challenge and setting new standards in detection performance.

Contribution

It presents a novel deep learning approach that combines multiple vision tasks within a single network and introduces an efficient multiscale sliding window method.

Findings

Won the ILSVRC2013 localization challenge

Achieved competitive detection and classification results

Established new state-of-the-art detection performance

Abstract

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.