Aggregated Residual Transformations for Deep Neural Networks

TLDR

Cardinality, the number of parallel transformations in a block, is introduced as a key architectural dimension alongside depth and width. The authors propose a simple, highly modularized network architecture for image classification. The architecture repeats a homogeneous multi‑branch block that aggregates transformations of identical topology, requiring only a few hyper‑parameters. On ImageNet‑1K, increasing cardinality improves accuracy while keeping complexity fixed, outperforming deeper or wider models; ResNeXt achieved second place in ILSVRC 2016 and outperformed ResNet on ImageNet‑5K and COCO detection. Code and pretrained models are publicly available online.

Abstract

We present a simple, highly modularized network architecture for image classification. Our network is constructed by repeating a building block that aggregates a set of transformations with the same topology. Our simple design results in a homogeneous, multi-branch architecture that has only a few hyper-parameters to set. This strategy exposes a new dimension, which we call cardinality (the size of the set of transformations), as an essential factor in addition to the dimensions of depth and width. On the ImageNet-1K dataset, we empirically show that even under the restricted condition of maintaining complexity, increasing cardinality is able to improve classification accuracy. Moreover, increasing cardinality is more effective than going deeper or wider when we increase the capacity. Our models, named ResNeXt, are the foundations of our entry to the ILSVRC 2016 classification task in which we secured 2nd place. We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its ResNet counterpart. The code and models are publicly available online.