ResNet, introduced by Kaiminh He, Xiangyu Zhang, Shaoquing Ren, and Jian Sun in the paper titled Deep Residual Learning for Image Recognition, was developed to address the accuracy degradation problem of deep neural networks with an increase in depth. This degradation is not caused by overfitting, but results from the fact that after some critical depth, the output looses the information of the input, so the correlation between the input and output starts diverging resulting in an increase in inaccuracy. The paper can be found at https://arxiv.org/abs/1512.03385.
ResNet-34 achieved a top-five validation error of 5.71%, better than BN-inception and VGG. ResNet-152 achieves a top-five validation error of 4.49%. An ensemble of six models with different depths achieved a top-five validation error of 3.57%, and won first place in ILSVRC-2015. ILSVRC stands for the...
