AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

AdaDistill: Adaptive Knowledge Distillation for Deep Face Recognition

Fadi Boutros, Vitomir Struc, Naser Damer

ECCV 2024 · Label-efficient and robust learning › Knowledge distillation · approach: Few-shot and metric distillation · 15 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
distill teacher class centers through margin softmax instead of vanilla MSE feature KD (ablation)MobileFaceNet student · IJB-C TAR@FAR=1e-5TAR (%): 79.79 -> 84.57Improved
adapt class centers with EMA instead of fixed teacher centers (ablation)ArcDistill · IJB-C TAR@FAR=1e-5TAR (%): 84.57 -> 88.23Improved
weight alpha by hard-sample importance (ablation)AdaArcDistill · IJB-C TAR@FAR=1e-5TAR (%): 88.23 -> 89.13Improved
use ArcFace margin 0.45 instead of 0.50 (ablation)AdaArcDistill · IJB-C TAR@FAR=1e-4TAR (%): 92.97 -> 93.27Improved
use CosFace instead of ArcFace margin penalty (ablation)AdaDistill · IJB-C TAR@FAR=1e-4TAR (%): 93.27 vs 93.05No clear effect
use stronger teacher ResNet100 instead of ResNet50 (ablation)AdaArcDistill · IJB-C TAR@FAR=1e-5TAR (%): 89.32 -> 87.48Hurt
use weaker teacher ResNet18 instead of ResNet50 (ablation)AdaArcDistill · IJB-C TAR@FAR=1e-5TAR (%): 89.32 -> 88.10Hurt
use a different student training dataset than the teacher (Casia-WebFace) (ablation)AdaArcDistill · IJB-C TAR@FAR=1e-5TAR (%): 57.75 -> 86.25Improved
use synthetic IDiff-Face as student training data (ablation)AdaArcDistill · IJB-C TAR@FAR=1e-5TAR (%): 3.43 -> 24.83Improved
use AdaArcDistill instead of ReFO+ on MegaFace and small benchmarksMobileFaceNet student · MegaFacerank-1 identification (%): AdaArcDistill 92.12 vs ReFO+ 92.41Mixed
use adaptive distillation instead of fixed centers (ablation)training cost · training time per batchseconds: 0.134 -> 0.247Hurt

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