AI Paper Atlas

Research map / Continual learning / Class-incremental

Class-Incremental Novel Class Discovery

Subhankar Roy, Mingxuan Liu, Zhun Zhong, Nicu Sebe, Elisa Ricci

ECCV 2022 · Continual learning › Class-incremental · approach: Exemplar-efficient class-incremental learning · 54 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 feature distillation and feature replay (ablation)FRoST · Average over three datasets (class-iNCD)All accuracy: 9.5 -> 58.3Improved
add feature replay (with feature distillation) (ablation)FRoST · Average over three datasets (class-iNCD)All accuracy: 17.4 -> 58.3Improved
add feature distillation (with feature replay) (ablation)FRoST · Average over three datasets (class-iNCD)All accuracy: 9.9 -> 58.3Improved
add self-training with joint classifier (ablation)FRoST · Average over three datasets (class-iNCD)All accuracy: 51.0 -> 58.3Improved
replace LwF (pre-softmax) with feature replay and distillation (ablation)FRoST · Average over three datasets (class-iNCD)All accuracy: 27.0 -> 58.3Improved
add feature replay to LwF (softmax) (ablation)LwF · Average over three datasets (class-iNCD)All accuracy: 21.1 -> 38.4Improved
add feature replay to LwF (pre-softmax) (ablation)LwF · Tiny-ImageNet (class-iNCD)Old accuracy: 6.8 -> 49.9Improved
add separate novel head to joint classifier (ablation)FRoST · Tiny-ImageNet (class-iNCD)New accuracy: 8.4 -> 33.9Improved
add novel head to joint classifier (ablation)FRoST · Average over three datasets (class-iNCD)All accuracy: 58.1 -> 58.3No clear effect
add self-training to joint-only classifier (ablation)Joint classifier · Average over three datasets (class-iNCD)All accuracy: 52.8 -> 58.1Improved
use FRoST instead of ResTuneclass-iNCD · Average over three datasets (class-iNCD)All accuracy: 48.3 -> 54.9Improved
use FRoST instead of ResTuneclass-iNCD · CIFAR-10 (class-iNCD)New accuracy: 0.0 -> 49.5Improved

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