Semi-Supervised ImageNet 10% Benchmark Added to Papers with Code
June 22, 2026
Papers with Code added a semi-supervised ImageNet benchmark tracking SOTA at 10% label usage, surfacing the S4L paper from 2019 as a historical baseline. The post highlights that LLM-era semi-supervised learning is rediscovering techniques vision researchers established years ago.
HOW THIS AFFECTS YOU
●
researcherUseful reference point if you're working on label-efficient training — the S4L baselines on 1% and 10% ImageNet remain strong historical anchors.