APEX: Assumption-free Projection-based Embedding eXamination Metric for Generative Distribution Evaluation
TL;DR
APEX (Assumption-free Projection-based Embedding eXamination), a novel evaluation framework leveraging two open-vocabulary foundation models as feature extractors and the Sliced Wasserstein Distance as similarity measure, proves that it inherits effective scalability to high-dimensional spaces.
Abstract
As generative models achieve unprecedented visual quality, the gold standard for image evaluation remains traditional feature-distribution metrics. However, these metrics are provably hindered by the closed-vocabulary bottleneck of outdated features and the assumptive bias of rigid parametric formulations. Recent alternatives exploit modern backbones to solve the feature bottleneck, yet continue to suffer from parametric limitations. To close this gap, we introduce APEX (Assumption-free Projection-based Embedding eXamination), a novel evaluation framework leveraging two open-vocabulary foundation models (CLIP and DINOv2) as feature extractors and the Sliced Wasserstein Distance (SWD) as similarity measure. Since SWD requires no assumptions about the compared data distributions, APEX operates as an assumption-free metric. Moreover, we prove with theoretical and empirical evidences that it inherits effective scalability to high-dimensional spaces. Benchmarking APEX metrics against established baselines, we show that they exhibit strong robustness to visual degradations, intra- and cross-dataset stability, ensuring highly stable evaluations on out-of-domain datasets.
