AI Paper Atlas

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

Masked Generative Distillation

Zhendong Yang, Zhe Li, Mingqi Shao, Dachuan Shi, Zehuan Yuan, Chun Yuan

ECCV 2022 · Label-efficient and robust learning › Knowledge distillation · approach: Knowledge distillation for detectors · 344 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 masked generation instead of direct mimicking (self-distillation) (ablation)ResNet-18 · ImageNetTop-1 accuracy: 70.05 -> 70.91Improved
mask random channels instead of spatial pixels (ablation)MGD ResNet-34 -> ResNet-18 · ImageNetTop-1 accuracy: 71.58 -> 71.69Improved
mask random channels instead of spatial pixels (ablation)MGD ResNet-50 -> MobileNet · ImageNetTop-1 accuracy: 72.35 -> 72.49Improved
use two 3x3 conv layers instead of one in generative block (ablation)MGD · ImageNetTop-1 accuracy: 71.28 -> 71.58Improved
use three conv layers instead of two in generative block (ablation)MGD · ImageNetTop-1 accuracy: 71.58 -> 71.49Hurt
use 5x5 kernel instead of 3x3 in generative block (ablation)MGD · ImageNetTop-1 accuracy: 71.58 -> 71.32Hurt
distill stage 4 instead of stage 1 (ablation)MGD ResNet-34 -> ResNet-18 · ImageNetTop-1 accuracy: 70.09 -> 71.58Improved
distill stages 2+3+4 together instead of stage 4 only (ablation)MGD · ImageNetTop-1 accuracy: 71.58 -> 71.47No clear effect
apply masking (lambda 0.5) instead of no mask (lambda 0) (ablation)MGD ResNet-34 -> ResNet-18 · ImageNetTop-1 accuracy: 71.41 -> 71.58Improved
stronger homogeneous teacher (ResNetV1D-152 vs ResNet-18) (ablation)MGD ResNet-18 student · ImageNetTop-1 accuracy: 70.91 -> 71.8Improved
heterogeneous ConvNeXt-T teacher instead of ResNet-34 (ablation)MGD ResNet-18 student · ImageNetTop-1 accuracy: 1.53 -> 0.88Hurt
add WSLD logit distillation to MGD (ablation)MGD ResNet-18 · ImageNetTop-1 accuracy: 71.58 -> 71.80Improved

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