Architecture
aka CNN
Convolutional Neural Network
Network using sliding convolution filters, dominant for vision before transformers.
Definition
CNNs exploit translation invariance with stacked convolution and pooling layers. ResNet, EfficientNet and ConvNeXt set image benchmarks for a decade. Vision transformers now match or exceed them at scale, but CNNs remain efficient for small models and on-device vision.
Common use cases
- Object detection
- Edge vision
- Medical imaging