AI Paper Atlas

Research map / Image generation / GANs

TREND: Truncated Generalized Normal Density Estimation of Inception Embeddings for GAN Evaluation

Junghyuk Lee, Jong-Seok Lee

ECCV 2022 · Image generation › GANs · approach: GAN latent optimization and evaluation · 5 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 truncated generalized normal instead of normal density (ablation)Inception feature density estimation · ImageNet Inception featuresaverage log-likelihood: -9463 -> 23867Improved
add truncation at zero to generalized normal (ablation)Inception feature density estimation · ImageNet Inception featuresaverage log-likelihood: 879 -> 23867Improved
add sharper-peak shape parameter to normal (untruncated) (ablation)Inception feature density estimation · ImageNet Inception featuresaverage log-likelihood: -9463 -> 879Improved
replace FID with TREND when sample count drops from 50000 to 5000BigGAN evaluation · BigGAN thresholds 0.4 and 0.6metric stability: FID 20.2 vs 21.1 (mean at 0.6 worse); TREND 0.0139 vs 0.0094Improved
use TREND instead of FID in a toy scenario with non-Gaussian densitiesToy GAN evaluation · Toy scenario 1ranking of model 2 vs 1: FID 0.0017 vs 0.0018; TREND 0.0086 vs 0.0022Improved
use TREND instead of FID when GAN fails to learn truthToy GAN evaluation · Toy scenario 2score: TREND 0.015 vs FID 1.4e-4Improved
increase image disturbance level 1 to 3 (Gaussian noise)TREND · ImageNet val disturbanceTREND: 0.0058 -> 0.0208Improved
increase random erasing level 1 to 3TREND · ImageNet val disturbanceTREND: 0.0369 -> 0.0827Improved
increase BigGAN truncation threshold 0.2 to infinityTREND · BigGAN ImageNetTREND: 0.0165 -> 0.0032Improved
increase BigGAN truncation threshold 0.2 to infinityIS · BigGAN ImageNetIS: 330.4 -> 144.6Hurt
use ADM-U instead of BigGAN (inf) under KIDKID · ImageNetKID: 0.0009 vs 0.0015Hurt

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