VGG-16

A foundational convolutional neural network that uses a deep stack of 3x3 filters to achieve high-precision image recognition.

Developed by the Visual Geometry Group at Oxford, VGG-16 is a 16-layer convolutional neural network that redefined computer vision through architectural simplicity. It utilizes a uniform sequence of small 3x3 receptive fields and 2x2 max-pooling layers to extract complex features from 224x224 input images. With approximately 138 million parameters and a 528 MB footprint, it secured a 92.7% top-5 accuracy on the ImageNet dataset during the 2014 ILSVRC competition. While newer models like ResNet offer better efficiency, VGG-16 remains a primary choice for transfer learning and feature extraction due to its predictable, linear structure.

https://arxiv.org/abs/1409.1556