Very Deep Convolutional Networks for Large-Scale Image Recognition

Andrew Zisserman; Karen Simonyan

doi:10.48550/arxiv.1409.1556

Very Deep Convolutional Networks for Large-Scale Image Recognition

Очень глубокие сверточные сети для масштабного распознавания изображений

Andrew Zisserman, Karen Simonyan

2014-09-04

SCID: 54.1/qv9azsyz Discuss with AI

16-19 weight layers3x3 convolution filtersImageNet Challenge 2014large-scale image recognitionvery deep convolutional networks

Abstract (AI)

In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Key Findings

Increasing convolutional network depth with very small 3x3 filters significantly improves large-scale image recognition accuracy.

Pushing network depth to 16–19 weight layers yields substantial improvement over prior-art configurations.

The learned deep representations generalise well to other datasets, achieving state-of-the-art results.

Their deep models achieved first and second places in ImageNet Challenge 2014 localisation and classification tracks respectively.

Two best-performing ConvNet models were made publicly available to facilitate further research.

Research Object

Very deep convolutional networks (ConvNets) with small (3x3) convolution filters trained for large-scale image recognition

Research Subject

Effect of network depth (increasing to 16–19 weight layers) on classification and localisation accuracy and the generalisation of learned visual representations to other image datasets

Publication Details

DOI

10.48550/arxiv.1409.1556

Publication Date

2014-09-04

Journal

Publisher

ISSN

Cited by

75434

Open access PDF

https://arxiv.org/pdf/1409.1556

Access Type

Author Information

Authors

Andrew Zisserman

Karen Simonyan

Download PDF

Make a presentation

Cited by20

Swin Transformer: Hierarchical Vision Transformer using Shifted Windows2021 Rethinking the Inception Architecture for Computer Vision2016 Squeeze-and-Excitation Networks2018 Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks2017 A survey on Image Data Augmentation for Deep Learning2019 Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network2017 Aggregated Residual Transformations for Deep Neural Networks2017 Non-local Neural Networks2018 Learning Deep Features for Discriminative Localization2016 ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices2018 Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising2017 ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks2020 Review of deep learning: concepts, CNN architectures, challenges, applications, future directions2021 Image Style Transfer Using Convolutional Neural Networks2016 Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning2016 Object Detection With Deep Learning: A Review2019 Least Squares Generative Adversarial Networks2017 A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects2021 Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions2021 Convolutional neural networks: an overview and application in radiology2018