AI Paper Atlas

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

A Fast Knowledge Distillation Framework for Visual Recognition

Zhiqiang Shen, Eric Xing

ECCV 2022 · Label-efficient and robust learning › Knowledge distillation · approach: Knowledge distillation for detectors · 56 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
use FKD soft labels instead of ReLabel label mapsResNet-50 from scratch · ImageNet-1KTop-1: 78.9 -> 79.8Improved
use FKD soft labels instead of ReLabel label mapsResNet-101 from scratch · ImageNet-1KTop-1: 80.7 -> 81.7Improved
train MEAL V2 with FKD instead of online teacher forwardsMEAL V2 ResNet-50 · ImageNet-1KTop-1: 80.67 -> 80.70No clear effect
use cosine lr and 1.5x epochs with FKD (ablation)MEAL V2 w/ FKD ResNet-50 · ImageNet-1KTop-1: 80.70 -> 80.91Improved
apply FKD to ViT-T without extra augmentation instead of DeiT KDViT-T · ImageNet-1KTop-1: 74.5 -> 75.2Improved
apply FKD to SReT-LT instead of its original KDSReT-LT · ImageNet-1KTop-1: 77.7 -> 78.7Improved
use FKD instead of S2-BNN for self-supervised distillationReActNet · ImageNet-1K linear evalTop-1: 61.5 -> 61.7No clear effect
use FKD instead of S2-BNN for self-supervised distillationResNet-50 · ImageNet-1K linear evalTop-1: 68.7 -> 68.8No clear effect
sample 4 crops per image in a mini-batch instead of 1 (ablation)vanilla ResNet-50 · ImageNet-1KTop-1: 77.18 -> 78.14Improved
sample 32 crops per image in a mini-batch instead of 1 (ablation)FKD ResNet-50 from scratch · ImageNet-1KTop-1: 79.59 -> 74.61Hurt
use full soft labels instead of hard labels (ablation)MEAL V2 w/ FKD ResNet-50 · ImageNet-1KTop-1: 80.20 -> 80.65Improved
use Marginal Smoothing (K=5) instead of Marginal Smoothing (K=10) (ablation)FKD from scratch ResNet-50 · ImageNet-1KTop-1: 79.44 -> 79.51No 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