Scalable Object Detection Using Deep Neural Networks

By Dumitru Erhan, Christian Szegedy, Alexander Toshev, Dragomir Anguelov. 2013

1,190 citations recorded by the source corpus.

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 cross-class 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.

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

Rethinking the Inception Architecture for Computer Vision (2015, methodology): “Also, improvements in the network quality resulted in new application domains for convolutional networks in cases where AlexNet features could not compete with hand engineered, crafted solutions, e.g. proposal generation in detection[4].”
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (2015, methodology): “The MultiBox methods [26], [27] generate region proposals from a network whose last fully-connected layer simultaneously predicts multiple class-agnostic boxes, generalizing the “single-box” fashion of OverFeat.”
Fast R-CNN (2015, methodology): “We note that [6] uses a related loss to train a class agnostic object proposal system.”
Going deeper with convolutions (2014, background): “The benefits of the architecture are experimentally verified on the ILSVRC 2014 classification and detection challenges, on which it significantly outperforms the current state of the art.”
End-to-End Object Detection with Transformers (2020, methodology): “re postprocessing-free. They need global inference schemes that model interactions between all predicted elements to avoid redundancy. For constant-size set prediction, dense fully connected networks [9] are sucient but costly. A general approach is to use auto-regressive sequence models such as rec”
Spatial Transformer Networks (2015, background): “In this section we discuss the prior work related to the paper, covering the central ideas of modelling transformations with neural networks [12, 13, 27], learning and analysing transformation-invariant representations [3, 5, 8, 17, 19, 25], as well as attention and detection mechanisms for feature ”
SSD: Single Shot MultiBox Detector (2015, methodology): “In the most recent works like MultiBox [2, 21], the selective search region proposals, which are based on low-level image features, are replaced by proposals generated directly from a separate deep neural network.”
Learning to decompose for object detection and instance segmentation (2015, background): “Most previous work on object detection first finds object candidates, and then determines the object category(Girshick et al., 2014; Erhan et al., 2014).”
You Only Look Once: Unified, Real-Time Object Detection (2015, methodology): “[8] In our tests, their most accurate model achieves 7 fps while a smaller, less accurate one runs at 18 fps.”
Towards Accurate Multi-person Pose Estimation in the Wild (2017, methodology): “Note that this approach is inspired by work on object detection, where a similar setup is used to predict bounding boxes, e.g. [14, 35].”
Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors (2016, methodology): “While both R-CNN and Fast R-CNN relied on an external proposal generator, recent works have shown that it is possible to generate box proposals using neural networks as well [40, 39, 8, 30].”
R-FCN: Object Detection via Region-based Fully Convolutional Networks (2016, background): “Another family of object detectors resort to fully-connected ( fc ) layers for generating holistic object detection results on an entire image, such as [26, 4, 18].”
Learning to Segment Object Candidates (2015, methodology): “Most closely related to our approach, Multibox [6, 30] proposed to train a ConvNet model to generate bounding box object proposals.”
What Makes for Effective Detection Proposals? (2015, methodology): “The simplest such approach would be to directly use the output of any hierarchical image segmentation algorithm, e.g. Gu et al. [34] use the segmentation produced by gPb [35].”
DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (2016, background): “DEEP Convolutional Neural Networks (DCNNs) [1] have pushed the performance of computer vision systems to soaring heights on a broad array of high-level problems, including image classification [2], [3], [4], [5], [6] and object detection [7], [8], [9], [10], [11], [12], where DCNNs trained in an end”
Pix2seq: A Language Modeling Framework for Object Detection (2021, methodology): “For example, the loss function is often based on bounding box regression (Szegedy et al., 2013; Lin et al., 2017b), intersection over union (Rezatoﬁghi et al., 2019), and set-based matching (Erhan et al., 2014; Liu et al., 2016; Redmon et al., 2016; Stewart et al., 2016; Carion et al., 2020).”
Focal Loss for Dense Object Detection (2017, methodology): “R-CNN was improved over the years, both in terms of speed [2], [29] and by using learned object proposals [3], [14], [21].”
Object-Proposal Evaluation Protocol is ‘Gameable’ (2015, background): “Types of Object Proposals: Object proposals can be broadly categorized into two categories: • Window scoring : In these methods, the space of all possible windows in an image is sampled to get a sub-set of the windows ( e.g ., via sliding window).”
Scalable, High-Quality Object Detection (2014, methodology): “The learning-based proposal method that our present work builds upon is the MultiBox approach presented in Erhan et al. [4], with improved underlying network architecture and training.”
DenseCap: Fully Convolutional Localization Networks for Dense Captioning (2015, methodology): “Further work has focused on processing all regions with only single forward pass of the CNN [17, 13], and on eliminating explicit region proposal methods by directly predicting the bounding boxes either in the image coordinate system [46, 9], or in a fully convolutional [31] and hence position-invar”