View source for VGGNet

{{Short description|Series of convolutional neural networks for image classification}}

{{Infobox software
| logo = 
| screenshot = 
| developer = Visual Geometry Group
| released = {{Start date and age|2014|09|04}}
| latest release version = 
| programming language = [[Caffe (software)]]
| engine = 
| platform = 
| genre = {{ indented plainlist |
*[[Convolutional neural network]]
*[[Deep neural network]]
}}
| license = [[CC BY 4.0]]
| website = {{URL|https://www.robots.ox.ac.uk/~vgg/research/very_deep/}}
}}
[[File:VGG_module_architecture.svg|thumb|VGG module architecture compared to [[AlexNet]] architecture.]]
The '''VGGNets''' are a series of [[convolutional neural network]]s (CNNs) developed by the Visual Geometry Group (VGG) at the [[University of Oxford]].

The VGG family includes various configurations with different depths, denoted by the letter "VGG" followed by the number of weight layers. The most common ones are VGG-16 (13 convolutional layers + 3 fully connected layers, 138M parameters) and VGG-19 (16 + 3, 144M parameters).<ref name=":1" />

The VGG family were widely applied in various computer vision areas.<ref>{{Cite journal |last1=Dhillon |first1=Anamika |last2=Verma |first2=Gyanendra K. |date=2020-06-01 |title=Convolutional neural network: a review of models, methodologies and applications to object detection |url=https://link.springer.com/article/10.1007/s13748-019-00203-0 |journal=Progress in Artificial Intelligence |language=en |volume=9 |issue=2 |pages=85–112 |doi=10.1007/s13748-019-00203-0 |issn=2192-6360}}</ref> An [[Ensemble learning|ensemble model]] of VGGNets achieved state-of-the-art results in the [[ImageNet Large Scale Visual Recognition Challenge]] (ILSVRC) in 2014.<ref name=":1">{{Citation |last1=Simonyan |first1=Karen |title=Very Deep Convolutional Networks for Large-Scale Image Recognition |date=2015-04-10 |last2=Zisserman |first2=Andrew|arxiv=1409.1556 }}</ref><ref>{{Cite web |title=ILSVRC2014 Results |url=https://image-net.org/challenges/LSVRC/2014/results.php |access-date=2024-09-06 |website=image-net.org}}</ref> It was used as a baseline comparison in the [[Residual neural network|ResNet]] paper for [[image classification]],<ref>{{Cite book |last1=He |first1=Kaiming |last2=Zhang |first2=Xiangyu |last3=Ren |first3=Shaoqing |last4=Sun |first4=Jian |chapter=Deep Residual Learning for Image Recognition |date=2016 |title=2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) |chapter-url=https://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html |pages=770–778|doi=10.1109/CVPR.2016.90 |arxiv=1512.03385 |bibcode=2016cvpr.confE...1H |isbn=978-1-4673-8851-1 }}</ref> as the network in the [[Region Based Convolutional Neural Networks|Fast Region-based CNN]] for [[object detection]], and as a base network in [[neural style transfer]].<ref>{{Cite conference |last1=Gatys |first1=Leon A. |last2=Ecker |first2=Alexander S. |last3=Bethge |first3=Matthias |date=2016 |title=Image Style Transfer Using Convolutional Neural Networks |url=https://www.cv-foundation.org/openaccess/content_cvpr_2016/html/Gatys_Image_Style_Transfer_CVPR_2016_paper.html |conference=IEEE Conference on Computer Vision and Pattern Recognition (CVPR) |pages=2414–2423}}</ref>

The series was historically important as an early influential model designed by composing generic modules, whereas [[AlexNet]] (2012) was designed "from scratch". It was also instrumental in changing the standard convolutional kernels in CNN from large (up to 11-by-11 in AlexNet) to just 3-by-3, a decision that was only revised in ConvNext (2022).<ref>{{Cite journal |last1=Liu |first1=Zhuang |last2=Mao |first2=Hanzi |last3=Wu |first3=Chao-Yuan |last4=Feichtenhofer |first4=Christoph |last5=Darrell |first5=Trevor |last6=Xie |first6=Saining |date=2022 |title=A ConvNet for the 2020s |journal=Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)|url=https://openaccess.thecvf.com/content/CVPR2022/html/Liu_A_ConvNet_for_the_2020s_CVPR_2022_paper.html |language=en |pages=11976–11986|arxiv=2201.03545 }}</ref><ref name=":0">{{Cite book |last1=Zhang |first1=Aston |title=Dive into deep learning |last2=Lipton |first2=Zachary |last3=Li |first3=Mu |last4=Smola |first4=Alexander J. |date=2024 |publisher=Cambridge University Press |isbn=978-1-009-38943-3 |location=Cambridge New York Port Melbourne New Delhi Singapore |chapter=8.2. Networks Using Blocks (VGG) |chapter-url=https://d2l.ai/chapter_convolutional-modern/vgg.html}}</ref>

VGGNets were mostly obsoleted by [[Inceptionv3|Inception]], [[Residual neural network|ResNet]], and [[DenseNet]]. RepVGG (2021) is an updated version of the architecture.<ref>{{Cite journal |last1=Ding |first1=Xiaohan |last2=Zhang |first2=Xiangyu |last3=Ma |first3=Ningning |last4=Han |first4=Jungong |last5=Ding |first5=Guiguang |last6=Sun |first6=Jian |date=2021 |title=RepVGG: Making VGG-Style ConvNets Great Again |journal=Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)|url=https://openaccess.thecvf.com/content/CVPR2021/html/Ding_RepVGG_Making_VGG-Style_ConvNets_Great_Again_CVPR_2021_paper.html |language=en |pages=13733–13742|arxiv=2101.03697 }}</ref>

== Architecture ==
[[File:Network-in-Network_architecture_vs_VGG_architecture.svg|thumb|Network-in-Network architecture compared to the VGG architecture. The Network in Network architecture (2013)<ref>{{Cite arXiv |eprint=1312.4400 |first1=Min |last1=Lin |first2=Qiang |last2=Chen |title=Network In Network |date=2013 |last3=Yan |first3=Shuicheng|class=cs.NE }}</ref> was an earlier CNN. It changed the AlexNet architecture by adding 1x1 convolutions, and using a global average pooling after the last convolution.]]The key architectural principle of VGG models is the consistent use of small <math>3 \times 3</math> convolutional filters throughout the network. This contrasts with earlier CNN architectures that employed larger filters, such as <math>11 \times 11</math> in AlexNet.<ref name=":0" />

For example, two <math display="inline">3 \times 3</math> convolutions stacked together has the same receptive field pixels as a single <math display="inline">5 \times 5</math> convolution, but the latter uses <math display="inline">\left(25 \cdot c^2\right)</math> parameters, while the former uses <math display="inline">\left(18 \cdot c^2\right)</math> parameters (where <math>c</math> is the number of channels). The original publication showed that deep and narrow CNN significantly outperform their shallow and wide counterparts.<ref name=":0" />

The VGG series of models are deep neural networks composed of generic modules:

# '''Convolutional modules''': <math>3 \times 3</math> convolutional layers with stride 1, followed by ReLU activations.
# '''Max-pooling layers''': After some convolutional modules, [[Pooling layer|max-pooling layers]] with a <math>2 \times 2</math> filter and a stride of 2 to downsample the feature maps. It halves both width and height, but keeps the number of channels.
# '''Fully connected layers''': Three fully connected layers at the end of the network, with sizes 4096-4096-1000. The last one has 1000 channels corresponding to the 1000 classes in ImageNet.
# '''Softmax layer''': A softmax layer outputs the probability distribution over the classes.

The VGG family includes various configurations with different depths, denoted by the letter "VGG" followed by the number of weight layers. The most common ones are VGG-16 (13 convolutional layers + 3 fully connected layers) and VGG-19 (16 + 3), denoted as configurations ''D'' and ''E'' in the original paper.<ref>{{Cite web |title=Very Deep Convolutional Networks for Large-Scale Visual Recognition |url=https://www.robots.ox.ac.uk/~vgg/research/very_deep/#pub |access-date=2024-09-06 |website=Computer Vision group from the University of Oxford}}</ref>

As an example, the 16 convolutional layers of VGG-19 are structured as follows:<math display="block">\begin{aligned}
&3 \to 64 \to 64 &\xrightarrow{\text{downsample}}\\
&64 \to 128 \to 128 &\xrightarrow{\text{downsample}}\\
&128 \to 256 \to 256 \to 256 \to 256 &\xrightarrow{\text{downsample}} \\
&256 \to 512 \to 512 \to 512 \to 512 &\xrightarrow{\text{downsample}}\\
&512 \to 512 \to 512 \to 512 \to 512 &\xrightarrow{\text{downsample}} 
\end{aligned}
</math>where the arrow <math>c_1 \to c_2</math> means a 3x3 convolution with <math>c_1</math> input channels and <math>c_2</math> output channels and stride 1, followed by ReLU activation. The <math>\xrightarrow{\text{downsample}}</math> means a down-sampling layer by 2x2 maxpooling with stride 2.
{| class="wikitable"
|+Table of VGG models
!Name
!Number of convolutional layers
!Number of fully connected layers
!Parameter count
|-
|VGG-16
|13
|3
|138M
|-
|VGG-19
|16
|3
|144M
|}

== References ==
<references />

[[Category:Deep learning software]]
[[Category:Object recognition and categorization]]
[[Category:Neural network architectures]]