Scalable Object Detection Using Deep Neural Networks

Abstract

Deep convolutional neural networks have recently achieved state-of-the-art performance on a number of image recognition benchmarks, including the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC-2012). The winning model on the localization sub-task was a network that predicts a single bounding box and a confidence score for each object category in the image. Such a model captures the whole-image context around the objects but cannot handle multiple instances of the same object in the image without naively replicating the number of outputs for each instance. In this work, we propose a saliency-inspired neural network model for detection, which predicts a set of class-agnostic bounding boxes along with a single score for each box, corresponding to its likelihood of containing any object of interest. The model naturally handles a variable number of instances for each class and allows for crossclass generalization at the highest levels of the network. We are able to obtain competitive recognition performance on VOC2007 and ILSVRC2012, while using only the top few predicted locations in each image and a small number of neural network evaluations.

DAPs: Deep Action Proposals for Action Understanding

“…achieving high object recall with less number of bounding boxes, preferably with a small computational overhead and the potential to scale to hundreds of object categories [35,40,37]. Here, discriminative methods based on deep learning models have helped improve the ranking quality of proposal approaches [7,37,28,32]. Inspired by this work, we extend the use of deep and recurrent networks to temporal action proposal generation by introducing a new architecture.…”

“…This problem can be solved by alternating between solving the assignment problem for a given θ k and back-propagating errors given an optimal assignment x k . For simplicity we rely on a heuristic similar to [7] to relax the assignment problem by introducing K anchor segments…”

Random clustering ferns for multimodal object recognition

“…This is especially critical for larger sizes of visual words, which are in turn need to obtain good clustering results. However, once RFCs are computed during the learning step, the proposed multimodal detector runs in about two frames per second using Matlab and MEX files 10 .…”

“…There exist a vast amount of methods that are able to detect and identify objects in images with striking results, in spite of diverse factors that make difficult this problem such M. Villamizar, A. Garrell, A. Sanfeliu and F. Moreno Institut de Robòtica i Informàtica Industrial, CSIC-UPC. E-mail: {mvillami,agarrell,sanfeliu,fmoreno}@iri.upc.edu as lighting changes, scaling, cluttered backgrounds, object deformations, general 3D rotations, and intra-class variations [10,11,15,26,38].…”

“…Machine learning techniques such as support vector machines [8,11,26,47], boosting [1,16,38,39], and more recently convolutional neural networks [10,15,23] are widely used to compute robust and discriminative classifiers for object detection. However, most of these works are computationally expensive and unfeasible for real-time applications.…”