AI Paper Atlas

Research map / Continual learning / Class-incremental

Online Continuous Generalized Category Discovery

Keon-Hee Park, Hakyung Lee, Kyungwoo Song, Gyeong-Moon Park

ECCV 2024 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention · 4 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 LoRA parameter-efficient tuning (ablation)fine-tuning baseline · CUB200 (OCGCD)forgetting F: 80.52 -> 22.24Improved
add energy-guided discovery (ablation)baseline + LoRA · CUB200 (OCGCD)M_new: 6.24 -> 23.67Improved
add energy-based contrastive loss (ablation)baseline + LoRA + EGD · CUB200 (OCGCD)M_all: 54.97 -> 57.37Improved
add variance-based feature augmentation (ablation)DEAN · CUB200 (OCGCD)M_new: 27.63 -> 33.11Improved
add VFA with K=5 augmented vectors (ablation)DEAN · CUB200 pseudo-labelsM_new^PS: 46.47 -> 65.75Improved
increase K above 5 in VFA (ablation)DEAN · CUB200 pseudo-labelsM_all^PS: 76.93 -> 75.44 (K=7), 75.22 (K=9)Hurt
use energy-guided discovery instead of CGCD fine-split (ablation)DEAN · CIFAR-100 (OCGCD)M_all: 52.27 -> 62.34Improved
use LoRA instead of Adapter or Prefix (ablation)DEAN · CUB200 (OCGCD)M_all: 56.96 (Adapter), 58.54 (Prefix) -> 59.56Improved
DEAN versus CGCD pseudo-label qualityDEAN · CUB200 pseudo-labelsM_all^PS: 61.88 -> 76.93Improved
add energy-based contrastive loss without VFA (ablation)DEAN w/o VFA · CUB200 pseudo-labelsM_new^PS: 45.80 -> 46.47No 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