# Resnet in Resnet: Generalizing Residual Architectures

**Authors:** Sasha Targ, Diogo Almeida, Kevin Lyman  
**arXiv:** [1603.08029](https://arxiv.org/abs/1603.08029)  
**Date:** 2016-03-25  
**Summary page:** https://www.summarizepaper.com/en/paper/1603.08029v1/

## Key points

- Authors introduce a new deep dual-stream architecture called Resnet in Resnet (RiR)
- RiR builds upon the success of residual networks (ResNets) and generalizes both ResNets and standard convolutional neural networks (CNNs)
- It can be easily implemented with no additional computational overhead
- RiR consistently outperforms ResNets on CIFAR-10 and establishes a new state-of-the-art on CIFAR-100
- RiR surpasses other architectures with similar amounts of augmentation
- The results suggest that RiR has the potential to improve performance on a wide range of computer vision tasks beyond those evaluated in this study.

## Summary

In their paper "Resnet in Resnet: Generalizing Residual Architectures," authors Sasha Targ, Diogo Almeida, and Kevin Lyman introduce a new deep dual-stream architecture called Resnet in Resnet (RiR). RiR builds upon the success of residual networks (ResNets) in computer vision tasks and generalizes both ResNets and standard convolutional neural networks (CNNs). Furthermore, it can be easily implemented with no additional computational overhead. The authors demonstrate that RiR consistently outperforms ResNets on CIFAR-10 and establishes a new state-of-the-art on CIFAR-100. Additionally, they show that RiR surpasses other architectures with similar amounts of augmentation. The results suggest that RiR has the potential to improve performance on a wide range of computer vision tasks beyond those evaluated in this study.