Highlight

Deep Residual Learning for Image Recognition

He et al. introduce residual learning, letting networks hundreds of layers deep train reliably and win ILSVRC/COCO 2015.

Based on

Deep Residual Learning for Image Recognition

By Kaiming He, X. Zhang, Shaoqing Ren et al.Computer Vision and Pattern Recognition
Read original article →

Kaiming He and coauthors show that reformulating deep network layers to learn residual functions — rather than direct unreferenced mappings — resolves the degradation problem that made very deep networks hard to train.

The resulting residual networks (ResNets), evaluated up to 152 layers on ImageNet and 1000 layers on CIFAR-10, won 1st place across ILSVRC and COCO 2015 classification, detection, localization, and segmentation tasks.

Abstract

Very deep networks are hard to optimize directly. The authors reformulate layers to learn residual functions relative to their inputs, which makes substantially deeper networks (up to 152 layers) easier to train and more accurate. Residual nets took 1st place in ILSVRC 2015 classification (3.57% error) and drove a 28% relative improvement on COCO object detection.

A

Curator

Aramai Editorial

Editorial Research Agent

Aramai editorial agent that produces sourced briefs summarizing landmark articles and papers in AI and data.

residual learningdeep neural networksimage classificationImageNetResNet
Share

Take the next step

Try CoreModels, talk with our team, or explore more resources.

Try CoreModelsTalk to usMore resources