Xception: Deep Learning with Depthwise Separable Convolutions

Chollet, François

doi:10.1109/cvpr.2017.195

Cited by 12,620 publications

(7,525 citation statements)

References 17 publications

Supporting

Mentioning

6,492

Contrasting

Unclassified

Order By: Relevance

“…This significantly decreases the number of parameters since the fully connected layers include a large number of parameters. Thus, this network is able to learn deeper representations of features with fewer parameters relative to AlexNet while it is much faster than VGG [31]. Figure 2 illustrates a compressed view of InceptionV3 employed in this study.…”

Section: Resnetmentioning

confidence: 99%

“…Xception network is similar to inception (GoogLeNet), wherein the inception module has been substituted with depth-wise separable convolutional layers [31]. Specifically, Xception's architecture is constructed based on a linear stack of a depth-wise separable convolution layer (i.e., 36 convolutional layers) with linear residual connections (see Figure 4).…”

Section: Xceptionmentioning

confidence: 99%

See 1 more Smart Citation

Very Deep Convolutional Neural Networks for Complex Land Cover Mapping Using Multispectral Remote Sensing Imagery

et al. 2018

View full text Add to dashboard Cite

Despite recent advances of deep Convolutional Neural Networks (CNNs) in various computer vision tasks, their potential for classification of multispectral remote sensing images has not been thoroughly explored. In particular, the applications of deep CNNs using optical remote sensing data have focused on the classification of very high-resolution aerial and satellite data, owing to the similarity of these data to the large datasets in computer vision. Accordingly, this study presents a detailed investigation of state-of-the-art deep learning tools for classification of complex wetland classes using multispectral RapidEye optical imagery. Specifically, we examine the capacity of seven well-known deep convnets, namely DenseNet121, InceptionV3, VGG16, VGG19, Xception, ResNet50, and InceptionResNetV2, for wetland mapping in Canada. In addition, the classification results obtained from deep CNNs are compared with those based on conventional machine learning tools, including Random Forest and Support Vector Machine, to further evaluate the efficiency of the former to classify wetlands. The results illustrate that the full-training of convnets using five spectral bands outperforms the other strategies for all convnets. InceptionResNetV2, ResNet50, and Xception are distinguished as the top three convnets, providing state-of-the-art classification accuracies of 96.17%, 94.81%, and 93.57%, respectively. The classification accuracies obtained using Support Vector Machine (SVM) and Random Forest (RF) are 74.89% and 76.08%, respectively, considerably inferior relative to CNNs. Importantly, InceptionResNetV2 is consistently found to be superior compared to all other convnets, suggesting the integration of Inception and ResNet modules is an efficient architecture for classifying complex remote sensing scenes such as wetlands.

show abstract

Section: Resnetmentioning

confidence: 99%

Section: Xceptionmentioning

confidence: 99%

Very Deep Convolutional Neural Networks for Complex Land Cover Mapping Using Multispectral Remote Sensing Imagery

et al. 2018

View full text Add to dashboard Cite

show abstract

“…Alternatively, one N×N convolution can be decomposed into two 1-D convolutions, one 1×N and one N×1 convolution [53]; this basically imposes a restriction that the 2-D filter must be separable, which is a common constraint in image processing [151]. Similarly, a 3-D convolution can be replaced by a set of 2-D convolutions (i.e., applied only on one of the input channels) followed by 1×1 3-D convolutions as demonstrated in Xception [152] and MobileNets [153]. The order of the 2-D convolutions and 1×1 3-D convolutions can be switched.…”

Section: Xmentioning

confidence: 99%

Efficient Processing of Deep Neural Networks: A Tutorial and Survey

et al. 2017

View full text Add to dashboard Cite

Abstract-Deep neural networks (DNNs) are currently widely used for many artificial intelligence (AI) applications including computer vision, speech recognition, and robotics. While DNNs deliver state-of-the-art accuracy on many AI tasks, it comes at the cost of high computational complexity. Accordingly, techniques that enable efficient processing of DNNs to improve energy efficiency and throughput without sacrificing application accuracy or increasing hardware cost are critical to the wide deployment of DNNs in AI systems.This article aims to provide a comprehensive tutorial and survey about the recent advances towards the goal of enabling efficient processing of DNNs. Specifically, it will provide an overview of DNNs, discuss various hardware platforms and architectures that support DNNs, and highlight key trends in reducing the computation cost of DNNs either solely via hardware design changes or via joint hardware design and DNN algorithm changes. It will also summarize various development resources that enable researchers and practitioners to quickly get started in this field, and highlight important benchmarking metrics and design considerations that should be used for evaluating the rapidly growing number of DNN hardware designs, optionally including algorithmic co-designs, being proposed in academia and industry.The reader will take away the following concepts from this article: understand the key design considerations for DNNs; be able to evaluate different DNN hardware implementations with benchmarks and comparison metrics; understand the trade-offs between various hardware architectures and platforms; be able to evaluate the utility of various DNN design techniques for efficient processing; and understand recent implementation trends and opportunities.

show abstract

“…Based on this premise, Chollet (2016) proposed a convolution performed independently over each channel of an input, followed by a pointwise convolution (i.e. a 1 × 1 convolution) projecting the channels output by depthwise convolution onto a new channel space.…”

Section: Xceptionmentioning

confidence: 99%

“…Xception stands for Extreme Inception and is the name of the architecture proposed by Chollet (2016).…”

Section: Xceptionmentioning

confidence: 99%

Fundus image analysis for automatic screening of ophthalmic pathologies.

Granero¹

View full text Add to dashboard Cite

AcknowledgmentsEste documento es el resultado de cuatro años de esfuerzo y dedicación que han propiciado mi evolución tanto a nivel profesional como personal. El principal motivo de dicho desarrollo han sido todas esas personas que han formado parte de este camino o que de alguna forma se han cruzado en el mismo. A todos ellos quiero dedicarles unas palabras de agradecimiento.En primer lugar, quiero agradecer a Valery, ella es la responsable de que yo esté ahora mismo escribiendo estas líneas. Me brindó la oportunidad de mi vida hace ya más de cinco años y me ha ido formando tanto en la vertiente investigadora como docente de la misma forma que una madre educa e inculca los valores esenciales a su hijo. Gracias por todo Valery, juntos cerramos una etapa clave en mi formación y juntos abrimos una nueva cargada de ilusión. Espero poder estar siempre a tu lado.También quiero agradecer todo el apoyo, energía y momentos divertidos que me han brindado mis compañeros de laboratorio. Gracias Sandra por todo lo que me has enseñado desde el primer momento en el que llegué hasta el día de hoy, no sé qué haría yo sin ti en esa sala. Gracias Fer, Félix, Andrés, Reinier, Ángel, Gabri, Fran P., Rober, Jorge y Adri por todos los consejos aportados y por amenizar como sólo vosotros sabéis mi día a día. También quiero recordar y agradecer a los que decidieron emprender otro rumbo: Javi, Conchi, Pablo, Eliseo, Fran M. y Miguel.De igual forma quiero mostrar mi gratitud hacia varias personas que han colaborado de una u otra forma en este trabajo. A Jesús Angulo, Débora Gil y Ramón Baldrich por su supervisión del trabajo desarrollado en las estancias llevadas a cabo durante esta etapa. A Kjersti Engan por aportar sus conocimientos sobre Dictionary Learning & Sparse Representation que han culminado en diversas publicaciones. A Rafael Molina por permitirme adentrarme en el fascinante mundo de los Procesos Gaussianos y sus múltiples aplicaciones.Gracias a mi madre Jacqueline, mi hermano Fran, mis abuelos Mª Carmen y Paco, mi padre Luis y Eva que siempre me han brindado su apoyo incondicional. Como bien sabéis sois fundamentales para mí, por ello ésta tesis también va dedicada a vosotros. No me puedo olvidar de Rosa que ocupa una parte grande de mi corazón. Os quiero.A Duke y Tom por todas esas horas de compañía desinteresada que me han ofrecido durante sus vidas y en especial en la época de redacción de este manuscrito.Por último, quiero dedicar especialmente este trabajo al motor de mi vida, a la persona que tiene que lidiar conmigo en los momentos más difíciles, a la persona que siempre me acaba sacando una sonrisa, a la persona que me enseña día tras día lo que es amor, fuerza, lucha, perseverancia e infinidad de cualidades que la definen. Gracias por todo cariño, sin tu apoyo estoy seguro de que esto no hubiera sido posible. Durante estos 10 años hemos abierto y cerrado muchas etapas juntos pero lo más ilusionante son todas aquellas que nos quedan por vivir. vi AbstractIn last years, the number of blindness cases has been significan...

show abstract

Xception: Deep Learning with Depthwise Separable Convolutions

Cited by 12,620 publications

References 17 publications

Very Deep Convolutional Neural Networks for Complex Land Cover Mapping Using Multispectral Remote Sensing Imagery

Very Deep Convolutional Neural Networks for Complex Land Cover Mapping Using Multispectral Remote Sensing Imagery

Efficient Processing of Deep Neural Networks: A Tutorial and Survey

Fundus image analysis for automatic screening of ophthalmic pathologies.

Contact Info

Product

Resources

About