AI Paper Atlas

Research map / Image generation / GANs

High-Fidelity GAN Inversion with Padding Space

Qingyan Bai, Yinghao Xu, Jiapeng Zhu, Weihao Xia, Yujiu Yang, Yujun Shen

ECCV 2022 · Image generation › GANs · approach: Large-scale GAN latent analysis · 27 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 padding coefficients at layers 0,1,3,5,7 on top of W+ inversion (ablation)pSp encoder · CelebA-HQ faceMSE: 0.0344 -> 0.0186Improved
add padding coefficients at layers 0,1,3,5,7 on top of W+ inversion (ablation)pSp encoder · LSUN ChurchLPIPS: 0.311 -> 0.209Improved
add padding for only the constant input p0 (ablation)pSp encoder · CelebA-HQ faceMSE: 0.0344 -> 0.0343No clear effect
extend padding to layer 9 (64x64) beyond layer 7 (ablation)PadInv · CelebA-HQ faceLPIPS: 0.090 -> 0.108Hurt
extend padding to layer 9 (64x64) beyond layer 7 (ablation)PadInv · LSUN ChurchMSE: 0.0838 -> 0.0979Hurt
add regularization loss L_reg on the inverted W+ code (ablation)PadInv · CelebA-HQ faceMSE: 0.0186 -> 0.0214Hurt
use PadInv instead of pSpencoder-based inversion · CelebA-HQ faceLPIPS: 0.16 -> 0.10Improved
use PadInv instead of optimization-based StyleGAN2 inversionStyleGAN2 inversion · LSUN ChurchMSE: 0.220 -> 0.086Improved
use PadInv instead of Restyle_pSp (reconstruction vs speed)iterative encoder · CelebA-HQ faceinference time (s): 0.2898 -> 0.0629Improved

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