AI Paper Atlas

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

One Coin Has Two Sides: Single Poistive Multi Label Learning from Salient Annotations

Xiaoyu Wang, Zhuoming Li, Bo Han, Hui LIU, Junhui Hou, Yuheng Jia

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Partial and multi-label learning

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 prototype-guided embedding loss (PL) (ablation)PiSA · SalSPML confidence-based, avg of COCO/NUS/VOCmAP: 59.49 -> 60.97Improved
add prototype-guided loss rejection (PGLR) (ablation)PiSA · SalSPML confidence-based, avg of COCO/NUS/VOCmAP: 59.49 -> 60.61Improved
add both PL and PGLR (ablation)PiSA · SalSPML confidence-based, avg of COCO/NUS/VOCmAP: 59.49 -> 62.63Improved
add PL and PGLR (ablation)PiSA · SalSPML confidence-based VOCmAP: 83.21 -> 87.12Improved
add PL and PGLR (ablation)PiSA · SalSPML confidence-based COCOmAP: 59.59 -> 63.43Improved
reject top-k unlabeled classes by prototype similarity instead of prediction probability (ablation)PiSA · SalSPML real VOC (k=3)mAP: +4.27%Improved
reject top-k unlabeled classes by prototype similarity instead of prediction probability (ablation)PiSA · SalSPML real COCO (k=3)mAP: +0.79%, +0.97%, +2.09%Improved
use adaptive threshold rejection (w=0.5, 1, 2) instead of fixed top-k=3 (ablation)PiSA · SalSPML real COCOmAP: 59.58 -> 59.21, 59.45, 58.97No clear effect
use adaptive threshold rejection (w=0.5, 1, 2) instead of fixed top-k=3 (ablation)PiSA · SalSPML real VOCmAP: 87.83 -> 87.32, 87.09, 85.53Hurt
apply PiSA instead of BCEBCE baseline · SalSPML real COCO annotation biasbias Delta_bias (lower better): 5.60 -> 4.57Improved
apply PiSA instead of BCEBCE baseline · SalSPML real VOC annotation biasbias Delta_bias (lower better): 3.40 -> 2.91Improved
apply PiSA instead of BCEBCE baseline · SalSPML confidence-based COCO and VOC annotation biasbias Delta_bias (lower better): 6.22 -> 4.49 (COCO), 4.10 -> 3.11 (VOC)Improved

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