VGG
VGG is a family of convolutional neural networks for images, created at Oxford in 2014, that demonstrated the value of depth by stacking small 3×3 filters. We explain its design idea, the VGG-16 and VGG-19 variants, its place in history—with the caveat that it is heavy and now surpassed—and its uses.
VGG is a family of convolutional neural network architectures for image classification, proposed in 2014 by Karen Simonyan and Andrew Zisserman, of the Visual Geometry Group at the University of Oxford, from which it takes its name. Its contribution was to show that increasing the network's depth improves accuracy while keeping a simple, regular design.
The idea: depth with small filters
The key to VGG is stacking many small, uniform 3×3 convolutions instead of using large filters. Stacking several 3×3 convolutions achieves the same receptive field as a larger filter, but with fewer parameters and more nonlinearities. Its two best-known variants differ in depth: VGG-16 (16 layers with weights) and VGG-19 (19 layers).
Its place in history
VGG was the basis of the Oxford team's entry in the 2014 ImageNet challenge, where it took first place in localization and second in classification. It was highly influential for its simplicity. A caveat is in order, however: it is a heavy network, with about 138 million parameters in VGG-16, computationally costly, and today surpassed in efficiency by later architectures such as ResNet.
What it is used for
Beyond classification, VGG is widely used for transfer learning: its features pre-trained on ImageNet serve as a starting point for other tasks, and it acts as the “backbone” of larger networks, for example in object detection or style transfer. It is distributed ready to use in frameworks such as PyTorch or Keras.
This article was produced with artificial intelligence under human editorial oversight.
