AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

S4L: Self-Supervised Semi-Supervised Learning

Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, Lucas Beyer

ICCV 2019 · Label-efficient and robust learning › Semi-supervised learning · approach: Safe and realistic semi-supervised learning · 780 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 S4L rotation self-supervised losssupervised ResNet50v2 · ILSVRC-2012 10% labelstop-5 accuracy (%): 80.43 -> 83.82Improved
add S4L rotation self-supervised losssupervised ResNet50v2 · ILSVRC-2012 1% labelstop-5 accuracy (%): 48.43 -> 53.37Improved
add S4L exemplar self-supervised losssupervised ResNet50v2 · ILSVRC-2012 1% labelstop-5 accuracy (%): 48.43 -> 47.02Hurt
add S4L exemplar self-supervised losssupervised ResNet50v2 · ILSVRC-2012 10% labelstop-5 accuracy (%): 80.43 -> 83.72Improved
add VAT losssupervised ResNet50v2 · ILSVRC-2012 1% labelstop-5 accuracy (%): 48.43 -> 44.05Hurt
add entropy minimization to VATVAT · ILSVRC-2012 10% labelstop-5 accuracy (%): 82.78 -> 83.39Improved
add pseudo-label retraining stepMOAM step 1 · ILSVRC-2012 10% labelstop-5 accuracy (%): 88.80 -> 89.96Improved
add final fine-tuning on the 10% labelsMOAM + pseudo label · ILSVRC-2012 10% labelstop-5 accuracy (%): 89.96 -> 91.23Improved
apply supervised loss on rotated copies of labeled images (ablation)S4L-Rotation · ILSVRC-2012top-5 accuracy (%): almost 1%Improved
apply self-supervised loss also on labeled images (ablation)S4L-Rotation · ILSVRC-2012top-5 accuracy (%)No clear effect
average predictions over 4 rotations at inference (ablation)S4L-Rotation · ILSVRC-2012top-5 accuracy (%): 1-2% more accurateImproved
fine-tune 4x-wider supervised vs add MOAM losses with all labelsResNet50v2 4x wider · ILSVRC-2012 100% labelstop-5 accuracy (%): 94.10 -> 94.97Improved

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