AI Paper Atlas

Research map / Continual learning / Class-incremental

Learnability-Driven Knowledge Assimilation for Class-Incremental Semantic Segmentation

Xinyue Zhang, Xu Zou, Wanjia Luo, Yanjie Wang, Jiahuan Zhou, Sheng Zhong, Luxin Yan

ICML 2026 · Continual learning › Class-incremental · approach: Unknown-class mining for incremental segmentation

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 Progressive Margin Learning (Lseg-PML) (ablation)baseline · Pascal VOC 15-1mIoU: 65.4 -> 77.2Improved
add Smooth Knowledge Distillation (LSKD) (ablation)baseline + PML · Pascal VOC 15-1mIoU: 77.2 -> 79.5Improved
add Misclassification-Aware Decoupling (LMAD) (ablation)baseline + PML + SKD · Pascal VOC 15-1mIoU: 79.5 -> 80.1Improved
use LMAD instead of standard contrastive learning (ablation)LDKA · Pascal VOC 15-1mIoU: 78.5 -> 79.5Improved
add competition matrix to contrastive learning (ablation)contrastive decoupling · Pascal VOC 15-1mIoU: 78.5 -> 78.8Improved
use LMAD instead of orthogonality constraints (ablation)LDKA · Pascal VOC 15-1mIoU: 77.4 -> 79.5Improved
raise margin/sensitivity ratio from balanced 1.0 to 1.5 (ablation)LDKA · Pascal VOC 15-1mIoU: 80.1 -> 79.1Hurt
lower margin/sensitivity ratio from 1.0 to 0.5 (ablation)LDKA · Pascal VOC 15-1mIoU: 80.1 -> 79.8Hurt
add LSKD to reduce inter-class confusion among classes 16-20 (ablation)LDKA · Pascal VOC 15-1 (confusion rate)mIoU: decreases of 7.5, 0.9, 5.4, 7.0, 2.0, 2.6, 4.4, 2.3, 0.6, and 10.9 percentage pointsImproved
LDKA versus MBS on ADE20K 100-5LDKA · ADE20K 100-5mIoU: 38.1 -> 40.1Improved
LDKA versus MBS on Pascal VOC 2-2LDKA · Pascal VOC 2-2mIoU: 72.6 -> 75.3Improved
LDKA versus MBS on Pascal VOC 10-1LDKA · Pascal VOC 10-1mIoU: 76.6 -> 76.7No clear effect

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