AI Paper Atlas

Research map / Image generation / GANs

EAGAN: Efficient Two-Stage Evolutionary Architecture Search for GANs

Guohao Ying, Xin He, Bin Gao, Bo Han, Xiaowen Chu

ECCV 2022 · Image generation › GANs · approach: Stabilized GAN training · 32 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 stage-2 discriminator search after generator search (ablation)EAGAN · CIFAR-10FID: 10.14 -> 9.91Improved
add stage-2 discriminator search after generator search (ablation)EAGAN · STL-10FID: 23.34 -> 22.18Improved
add stage-2 discriminator search after generator search (ablation)EAGAN · STL-10IS: 10.02 -> 10.44Improved
use simple conv initial D in stage-1 instead of the AdversarialNAS discriminator (ablation)EAGAN · CIFAR-10FID: 9.91 -> 12.84Mixed
repeat two-stage search with previous D as initial D (ablation)EAGAN · CIFAR-10FID: 9.91 -> 10.53No clear effect
search G first rather than D first (ablation)EAGAN · CIFAR-10gradient stabilityImproved
decouple G and D search into two stages instead of coupled evolution (ablation)EAGAN · CIFAR-10search stabilityImproved
add weight-resetting in stage-2 (ablation)EAGAN · CIFAR-10IS and FID during stage-2Improved
use EAGAN-G instead of AdversarialNAS (G only)generator search · CIFAR-10FID: 24.04 -> 10.14Improved
use EAGAN-GD1 instead of AlphaGAN when searching G and DGAN NAS · CIFAR-10FID: 15.56 -> 9.91Improved

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