In simple words: VGG uses repeated small filters in a deep CNN.
Repeated small filters
Visual Geometry Group (VGG) models stack many small 3 by 3 convolution filters. Several small filters in sequence can build richer patterns while keeping a regular design.
Trade-off
Deeper layers can progress from edges to object parts. VGG models are relatively large and computationally heavy, so later architectures looked for better efficiency.
See the flow
3×3 filters→deeper features→image class
Simple example
To recognise a dog, early layers detect edges, middle layers detect fur and ears, and later layers combine them into a whole animal.
Remember: VGG uses repeated small filters in a deep CNN.