AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Regularizing Deep Networks with Semantic Data Augmentation

Yulin Wang, Gao Ming Huang, Shiji Song, Xuran Pan, Yitong Xia, Cheng Wu

TPAMI 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 179 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add ISDA to trainingResNet-50 · ImageNet valtop-1 error (%): 23.0 -> 21.9Improved
add ISDA to trainingDenseNet-BC-121 · ImageNet valtop-1 error (%): 23.7 -> 23.2Improved
add ISDA to AutoAugmentShake-Shake (26, 2x112d) · CIFAR-100test error (%): 15.21 -> 13.87Improved
add ISDA to VATCNN-13 · CIFAR-10 (1,000 labels)test error (%): 18.12 -> 14.38Improved
add ISDA to PSPNetPSPNet ResNet-101 · Cityscapes valmIoU (%): 77.46 -> 78.72Improved
use dynamically estimated per-class covariance instead of identity matrix (ablation)ISDA · CIFAR-100test error (%): 18.53 -> 17.98Improved
use full covariance instead of diagonal matrix (ablation)ISDA · CIFAR-100 + Cutouttest error (%): 17.54 -> 16.95Improved
use per-class covariance instead of a single global covariance (ablation)ISDA · CIFAR-100 + Cutouttest error (%): 18.12 -> 16.95Improved
use iteration-dependent lambda_0 instead of constant lambda_0 (ablation)ISDA · CIFAR-100 + Cutouttest error (%): 17.34 -> 16.95Improved
increase lambda_0 to 10 (ablation)ISDA · CIFAR-100 sensitivitytest errorHurt
use ISDA upper bound (M to infinity) instead of sampled explicit augmentation with small M (ablation)explicit SDA · CIFAR-100 + Cutout / AutoAugmenttest errorImproved
use ISDA instead of GAN-based augmentationResNet-110 · CIFAR-100test error (%): 27.57 -> 28.25Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy