,

Abstract

We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously.

Key Contributions

  1. Residual Blocks: Skip connections that enable gradient flow
  2. ResNet-50/101/152: Architectures that won ILSVRC 2015
  3. Enabling Deep Learning: Made 100+ layer networks trainable

Results

Architecture Top-1 Error Top-5 Error
VGG-16 28.5 9.9
ResNet-152 21.7 5.9