CNN architectures: AlexNet to MobileNet · MAAI7602 Applied Computer Vision

Classic CNNs differ in how they spend parameters and computation: large early filters (AlexNet), stacks of 3 × 3 filters (VGG), parallel branches with 1 × 1 reductions (Inception) and depthwise separable filters (MobileNet). Which design choices distinguish the major CNN architectures, and how is their cost measured?

This page needs JavaScript for its explanations, activities and practice questions.