AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Privacy Assessment on Reconstructed Images: Are Existing Evaluation Metrics Faithful to Human Perception?

Xiaoxiao Sun, Nidham Gazagnadou, Vivek Sharma, Lingjuan Lyu, Hongdong Li, Liang Zheng

NeurIPS 2023 · Image generation › Autoregressive and latent generation · approach: Learning-based image quality assessment · 2 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 SemSim instead of PSNRprivacy metric · CIFAR-100Kendall tau with human ranking: 0.4725 -> -0.7143Improved
use SemSim instead of best baseline FIDprivacy metric · CIFAR-100Kendall tau: -0.5604 -> -0.7143Improved
use SemSim instead of best baseline MSEprivacy metric · Caltech-101Kendall tau: -0.5525 -> -0.6889Improved
use SemSim instead of best baseline LPIPSprivacy metric · ImagenetteKendall tau: -0.5922 -> -0.7012Improved
use SemSim instead of best baseline PSNRprivacy metric · CelebAKendall tau: 0.5604 -> -0.6923Improved
use SemSim instead of best baseline SSIM on fine-grained datasetprivacy metric · Stanford DogsKendall tau: 0.3048 -> -0.5938Improved
evaluate SemSim trained on InvGrad against unseen DLG attackSemSim · CIFAR-100Kendall tau: -0.5237 -> -0.7342Improved
evaluate SemSim on unseen CAFE attackSemSim · CIFAR-100Kendall tau: 0.5392 -> -0.7510Improved
evaluate SemSim on unseen GradAttackSemSim · CIFAR-100Kendall tau: -0.4819 -> -0.7260Improved
reduce number of human-annotated training samples down to 50 (ablation)SemSim · CIFAR-100correlation with human: 50 samplesMixed
use simple LeNet backbone instead of ResNet50 (ablation)SemSim · CIFAR-100Kendall tau: 0.65 vs 0.5604Mixed
triplet loss instead of contrastive loss (ablation)SemSim · CIFAR-100correlation with humanNo clear effect

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