CNN Architectures

← Back to Convolutional Neural Networks

The evolution of CNN designs from simple to deep and efficient. Each architecture introduced key innovations that advanced the field.

Key Architectures

• LeNet — original CNN (Yann LeCun, 1998), digit recognition
• AlexNet — deep CNN breakthrough on ImageNet (2012), ReLU, dropout
• VGG — very deep with small 3x3 filters (2014)
• ResNet — residual connections enabling 100+ layer networks (2015)
• EfficientNet — compound scaling of depth, width, and resolution (2019)
• ConvNeXt — modernized CNN competing with Vision Transformers (2022)

Related

• Residual Connections (ResNet innovation)
• Vision Transformers (modern alternative to CNNs)

---

deep-learning cnn architectures