AI Paper Atlas

Research map / Image generation / GANs

Loss-Sensitive Generative Adversarial Networks on Lipschitz Densities

Guo-Jun Qi

IJCV 2019 · Image generation › GANs · approach: Stabilized GAN training · 368 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 CLS-GAN loss-sensitive objective instead of DCGAN featuresCLS-GAN · CIFAR-10 (all labels)accuracy: 82.8% -> 91.7%Improved
use CLS-GAN instead of Improved GAN with 400 labels per classCLS-GAN · CIFAR-10 (400 labels per class)accuracy: 81.4% -> 82.7%Improved
use CLS-GAN instead of ALI on SVHNCLS-GAN · SVHN (1000 labels)error rate: 7.3% -> 5.98%Improved
remove batch normalization from generator (ablation)DCGAN · CelebA generationimage qualityHurt
remove batch normalization from generator (ablation)LS-GAN · CelebA generationimage qualityNo clear effect
update generator every 3 or 5 iterations while loss updated every iteration (ablation)LS-GAN · CelebA generationimage qualityNo clear effect
use L1 distance of image features as margin; use external Inception/VGG-16 features (ablation)LS-GAN · CelebA generationimage qualityImproved
halve lambda (ablation)CLS-GAN · MNIST, CIFAR-10, SVHN generationsample diversityHurt
use regularized GAN (WGAN) instead of DCGANWGAN · CIFAR-10 test MREMRE: 0.1506 -> 0.1109Improved
use GLS-GAN instead of DCGANGLS-GAN · CIFAR-10 test MREMRE: 0.1506 -> 0.1089Improved
use GLS-GAN with LeakyReLU slope 0.5 on tiny ImageNetGLS-GAN · tiny ImageNet test MREMRE: 0.2085Improved

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