AI Paper Atlas

Research map / Continual learning / Class-incremental

Background Adaptation with Residual Modeling for Exemplar-Free Class-Incremental Semantic Segmentation

Anqi Zhang, Guangyu Gao

ECCV 2024 · Continual learning › Class-incremental · approach: Unknown-class mining for incremental segmentation · 13 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 Background Adaptation residual channel instead of shared background classifier (ablation)BAR baseline · Pascal VOC 15-1mIoU all: 61.1 -> 66.3Improved
add Background Feature Distillation (ablation)BAR · Pascal VOC 15-1mIoU novel: 30.5 -> 37.1Improved
add Group Knowledge Distillation on top of BFD (ablation)BAR · Pascal VOC 15-1mIoU all: 67.3 -> 67.6No clear effect
add triplet background adaptation loss BgA- (ablation)BAR · Pascal VOC 15-1mIoU novel: 36.5 -> 43.4Improved
add BgA+ loss without BgA- (ablation)BAR · Pascal VOC 15-1mIoU all: 67.6 -> 66.5Hurt
combine BgA+ and BgA- losses (ablation)BAR · Pascal VOC 15-1mIoU all: 69.0 -> 70.0Improved
apply Filter operation that zeroes positive adaptation logits (ablation)BAR · Pascal VOC 15-1mIoU all: 60.8 -> 70.0Improved
use Background Feature Distillation instead of MSE over all regions (ablation)BAR · Pascal VOC 15-1mIoU novel: 44.3 -> 45.9Improved
use triplet loss for background region instead of BCE-1 (ablation)BAR · Pascal VOC 15-1mIoU all: 69.6 -> 70.0Improved
use BAR vs DKD on short-term ADE20K tasksBAR · ADE20K 100-50mIoU all: 36.0 -> 35.7Hurt

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