Revisiting ResNets: Improved Training and Scaling Strategies

TL;DR

This paper revisits ResNet architectures, demonstrating that training and scaling strategies are more impactful than architecture changes, leading to faster, high-performing models suitable as strong baselines.

Contribution

It introduces new scaling strategies and a family of ResNet-RS architectures that outperform or match state-of-the-art models in speed and accuracy.

Findings

ResNet training and scaling strategies are more influential than architecture modifications.

ResNet-RS models are 1.7x - 2.7x faster than EfficientNets on TPUs with similar accuracy.

ResNet-RS achieves 86.2% top-1 accuracy on ImageNet and improves transfer learning performance.

Abstract

Novel computer vision architectures monopolize the spotlight, but the impact of the model architecture is often conflated with simultaneous changes to training methodology and scaling strategies. Our work revisits the canonical ResNet (He et al., 2015) and studies these three aspects in an effort to disentangle them. Perhaps surprisingly, we find that training and scaling strategies may matter more than architectural changes, and further, that the resulting ResNets match recent state-of-the-art models. We show that the best performing scaling strategy depends on the training regime and offer two new scaling strategies: (1) scale model depth in regimes where overfitting can occur (width scaling is preferable otherwise); (2) increase image resolution more slowly than previously recommended (Tan & Le, 2019). Using improved training and scaling strategies, we design a family of ResNet architectures, ResNet-RS, which are 1.7x - 2.7x faster than EfficientNets on TPUs, while achieving similar accuracies on ImageNet. In a large-scale semi-supervised learning setup, ResNet-RS achieves 86.2% top-1 ImageNet accuracy, while being 4.7x faster than EfficientNet NoisyStudent. The training techniques improve transfer performance on a suite of downstream tasks (rivaling state-of-the-art self-supervised algorithms) and extend to video classification on Kinetics-400. We recommend practitioners use these simple revised ResNets as baselines for future research.