D&amp;#x00E9;j&amp;#x00E0; Vu object localization using IRF neural networks properties

Smagghe, Philippe; Buessler, Jean-Luc; Urban, Jean-Philippe

doi:10.1109/ijcnn.2013.6706916

Cited by 3 publications

(5 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unlike conventional Feedforward Networks, the IRF-NN uses the input images directly without prior feature extraction (Smagghe et al, 2013). Furthermore, the hidden units are partially connected to all the previous units, instead of being totally connected to these nodes (Fig.…”

Section: Receptive Fields Based Architecturesmentioning

confidence: 99%

“…The activation function used for the output is a sigmoid function (1). Where i is the neuron and x, y are the coordinates of the image pixels (Smagghe et al, 2013 IRF-NN is a Feedforward Network that is more suitable for image recognition. It ensures a fast and accurate classification, by dealing directly with original images and without the need of a preprocessing stage to extract features from raw data.…”

Section: Receptive Fields Based Architecturesmentioning

confidence: 99%

See 1 more Smart Citation

RF-CNNS: Thin Deep Learning Networks For Accelerating Traffic Signs Recognition

Bousarhane,

Bouzidi

2023

Journal of Computer Science

View full text Add to dashboard Cite

Road scene analysis is a wide domain of research that aims to ameliorate the environmental perception in intelligent transportation systems, including autonomous vehicles and advanced driver-assistance systems. It also plays a crucial role in road safety improvement, by contributing to the reduction of traffic accidents' rate. Ensuring an efficient recognition of traffic signs contributes enormously to making cars safer, which in consequence helps to save more lives on roads. To ensure this recognition, many approaches are adopted, especially deep learning ones and more specifically convolutional neural networks. In effect, these networks have proven their high performances in many fields of computer vision research, including traffic sign recognition. However, although their high performances, many limitations still face their implementation, especially in real-time applications and resource-constrained environments. From this perspective, creating a certain balance between the model complexity and the classification accuracy, through a computationally efficient network, is the main objective of this study. To achieve this goal, a receptive fields architecture is adopted to preserve and optimize the connectivity between the different units of the network. Based on this architecture, two receptive field networks are proposed, with reduced complexity and enhanced generalizability. Using two public datasets, the obtained results show that the adopted approach ensures high classification accuracies and considerably accelerates the inference stage. The obtained accuracy is about 98.49%, using the Belgium traffic signs classification dataset, while the inference time is less than 500 us per image.

show abstract

Section: Receptive Fields Based Architecturesmentioning

confidence: 99%

Section: Receptive Fields Based Architecturesmentioning

confidence: 99%

RF-CNNS: Thin Deep Learning Networks For Accelerating Traffic Signs Recognition

Bousarhane,

Bouzidi

2023

Journal of Computer Science

View full text Add to dashboard Cite

show abstract

“…This section presents and illustrates some IRF-NN properties. A few of these properties and perspectives of application have been presented in recent conference papers [5,6,7]. We also carried out an in-depth study to determine the parameters of the network, its encoding performances on large image datasets, and its generalization capabilities.…”

Section: Network Properties and Applicationsmentioning

confidence: 99%

“…The IRF-NN has an interesting supplementary property: it can distinguish a known image from an unknown one [7]. A novel image induces a network response that differs significantly from the one observed for views similar to the learning set.…”

Section: Novelty Detectionmentioning

confidence: 99%

See 1 more Smart Citation

Image receptive fields for artificial neural networks

Buessler¹,

Smagghe²,

Urban³

2014

Neurocomputing

Self Cite

View full text Add to dashboard Cite

This article describes the structure of the Image Receptive Field Neural Network (IRF-NN), introduced recently by our team. This structure extends simplified learning introduced by Extreme Learning Machine and Reservoir Computing techniques to the field of images. Neurons are organized in a single hidden layer feedforward network architecture with an original organization of the network's input weights. To represent color images efficiently, without prior feature extraction, the weight values linked to a neuron are determined by a 2-D Gaussian function. The activation of a neuron by an image presents the properties of a nonlinear localized receptive field, parameterized with a small number of degrees of freedom. This article shows that an efficient representation of the images is provided by a large number of neurons, each associated to a randomly initialized and constant receptive field. The training step determines only the output weights of the network. It is therefore extremely fast, without retropropagation or iterations, and remains efficient with large sets of images. The network is easy to implement, presents excellent generalization performances for classification applications, and allows the detection of unknown inputs. The efficiency of this technique is illustrated with several benchmarks, photo and video datasets.

show abstract

View and Illumination Invariant Object Classification Based on 3D Color Histogram Using Convolutional Neural Networks

Ijjina

Mohan

2015

Computer Vision - ACCV 2014 Workshops

View full text Add to dashboard Cite

Déjà Vu object localization using IRF neural networks properties

Cited by 3 publications

References 13 publications

RF-CNNS: Thin Deep Learning Networks For Accelerating Traffic Signs Recognition

RF-CNNS: Thin Deep Learning Networks For Accelerating Traffic Signs Recognition

Image receptive fields for artificial neural networks

View and Illumination Invariant Object Classification Based on 3D Color Histogram Using Convolutional Neural Networks

Contact Info

Product

Resources

About