ResNet won the 2015 ImageNet Large Scale Visual Recognition Challenge. Developed by researchers at Microsoft Research, ResNet introduced residual learning, allowing networks to learn corrections relative to their input rather than fitting every transformation directly.
Its defining feature is the shortcut, or skip, connection. These connections help information and gradients move through very deep networks, reducing the degradation problem that had made simply adding layers unreliable. The winning system was a 152-layer model, much deeper than many earlier vision networks.
ResNet is sometimes confused with AlexNet, which won ImageNet in 2012 and helped spark the modern deep-learning boom. ResNet did not merely win because it was large; its residual architecture made depth more practical. Later versions, including ResNet-50 and ResNet-101, became widely used as computer-vision backbones.