AI Paper Atlas

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

Virtual Category Learning: A Semi-Supervised Learning Method for Dense Prediction With Extremely Limited Labels

Changrui Chen, Jungong Han, Kurt Debattista

TPAMI 2024 · Label-efficient and robust learning › Semi-supervised learning · approach: Consistency training for semi-supervised segmentation · 94 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 VC learning to teacher-student baselineteacher-student baseline · Pascal VOC 1/128 (mean of 3 folds)mIoU: 50.77 -> 53.00Improved
add VC learning (ResNet101 backbone)teacher-student baseline · Pascal VOC 1/128, fold 2mIoU: 50.79 -> 55.23Improved
add VC learningteacher-student baseline · Cityscapes 1/72 (mean)mIoU: 51.18 -> 53.02Improved
use CE instead of MSE loss in VC (ablation)VC learning · Pascal VOC 1/64 (mean)mIoU: 62.20 -> 62.78Improved
add virtual category to MSE loss (vs NEG only) (ablation)VC learning · Pascal VOC 1/64 (mean)mIoU: 61.71 -> 62.20Improved
use virtual weight with magnitude (VC) instead of cosine similarity (ablation)VC learning · Pascal VOC 1/128 fold 1mIoU: 47.76 -> 48.62Improved
replace discarding confusing samples with VC (ablation)VC learning · MS COCO 1%mAP: 19.37 -> 20.81Improved
replace omit-and-detach with VC (ablation)VC learning · MS COCO 1%mAP: 19.70 -> 20.81Improved
keep all potential labels for confusing samples (ablation)pseudo-labelling baseline · MS COCO 1%mAP: 20.00 -> 19.36Hurt
add Reg* localisation loss on top of VC (ablation)VC learning · MS COCO 1%mAP: 20.81 -> 20.94Improved
use cross-model verification instead of temporal stability (ablation)VC learning · MS COCO 1%mAP: 20.81 -> 20.96Improved
use strong augmentation for virtual weight generation (ablation)VC learning · MS COCO 1%mAP: 20.81 -> 20.70Hurt

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