All terms
Architectures
Residual Network
A deep image network built on skip connections that add each block's input to its output.
Definition
A Residual Network, or ResNet, is a convolutional architecture built around skip connections that add a block's input directly to its output. This identity shortcut lets gradients flow cleanly and allows training to hundreds of layers without the accuracy degradation that plagued earlier deep networks. ResNet became a standard backbone for image classification and detection, and its residual idea later carried into Transformers.