AI Paper Atlas

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

Mind the Gap in Distilling StyleGANs

Guodong Xu, Yuenan Hou, Ziwei Liu, Chen Change Loy

ECCV 2022 · Label-efficient and robust learning › Knowledge distillation · approach: Few-shot and metric distillation · 16 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
inherit teacher mapping network weights (ablation)StyleGAN2 student distillation · FFHQ 256x256 StyleGAN2 ablationFID: 11.27 -> 8.23Improved
inherit teacher convolution weights in addition to mapping network (ablation)StyleGAN2 student distillation · FFHQ 256x256 StyleGAN2 ablationFID: 7.94 -> 8.45Hurt
add RGB+LPIPS mimicking when mapping network is randomly initialized (ablation)StyleGAN2 student distillation · FFHQ 256x256 StyleGAN2 ablationFID: 10.92 -> 11.27Hurt
add RGB+LPIPS mimicking when mapping network is inherited (ablation)StyleGAN2 student distillation · FFHQ 256x256 StyleGAN2 ablationFID: 10.54 -> 8.61Improved
add latent-direction (LD) relation loss (ablation)StyleGAN2 student (inherit mapping) · FFHQ 256x256 StyleGAN2 ablationFID: 8.23 -> 7.94Improved
inherit mapping network instead of random init (ablation)StyleGAN2 student (random conv) · FFHQ 256x256 mapping network studyFID: 11.78 -> 8.30Improved
two-stage mapping mimicking with L1 instead of random init (ablation)StyleGAN2 student · FFHQ 256x256 mapping network studyFID: 11.78 -> 9.69Improved
use L1 instead of L2 for mapping mimicking (ablation)two-stage mapping training · FFHQ 256x256 mapping network studyFID: 10.80 -> 9.69Improved
use latent-direction augmentation instead of random offset (ablation)relation distillation · FFHQ 256x256 relation ablationFID: 9.80 -> 9.05Improved
use relation between normal and LD-augmented samples instead of single view (ablation)relation distillation · FFHQ 256x256 relation ablationFID: 9.47 -> 9.05Improved
use KL instead of L2 for LD loss with RGB+LPIPS (ablation)LD loss · FFHQ 256x256 relation ablationFID: 8.64 -> 8.26Improved
use proposed method instead of CAGANStyleGAN2 distillation · FFHQ 256x256FID: 7.9 -> 7.25Improved

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