AI Paper Atlas

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

Semi-Supervised Semantic Segmentation With Pixel-Level Contrastive Learning From a Class-Wise Memory Bank

Iñigo Alonso, Alberto Sabater, David Ferstl, Luis Montesano, Ana C. Murillo

ICCV 2021 · Label-efficient and robust learning › Semi-supervised learning · approach: Consistency training for semi-supervised segmentation · 288 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 contrastive loss to supervised loss alone (ablation)Teacher-student segmentation · Cityscapes val (1/30 labels, ablation)mIoU: 49.5 -> 54.4Improved
add pseudo-label loss (ablation)Teacher-student segmentation · Cityscapes val (1/30 labels, ablation)mIoU: 49.5 -> 56.7Improved
add entropy minimization loss to supervised loss (ablation)Teacher-student segmentation · Cityscapes val (1/30 labels, ablation)mIoU: 49.5 -> 52.2Improved
add contrastive loss to sup + pseudo + ent (ablation)Full method · Cityscapes val (1/30 labels, ablation)mIoU: 57.4 -> 59.4Improved
add entropy loss to sup + pseudo + contr (ablation)Full method · Cityscapes val (1/30 labels, ablation)mIoU: 59.0 -> 59.4Improved
add pseudo-label loss to sup + ent + contr (ablation)Full method · Cityscapes val (1/30 labels, ablation)mIoU: 57.3 -> 59.4Improved
set contrastive weight to 0.1 instead of 10^4 (ablation)Contrastive loss · Cityscapes val (1/30 labels, ablation)mIoU: 50.3 -> 59.4Improved
set contrastive weight to 10^-4 instead of 0.1 (ablation)Contrastive loss · Cityscapes val (1/30 labels, ablation)mIoU: 59.4 -> 57.6Hurt
increase memory bank size per class from 32 to 256 (ablation)Memory bank · Cityscapes val (1/30 labels, ablation)mIoU: 58.7 -> 59.4Improved
increase memory bank size from 256 to 512 (ablation)Memory bank · Cityscapes val (1/30 labels, ablation)mIoU: 59.4 -> 59.3No clear effect
add Feature Quality Filter to base contrastive module (ablation)Contrastive module · Cityscapes val (1/30 labels, ablation)mIoU: 58.3 -> 59.0Improved
extract memory features with teacher (ablation)Contrastive module · Cityscapes val (1/30 labels, ablation)mIoU: 58.3 -> 58.7Improved

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