Modelpedia
Work in progress
About
Findings
Models
Concepts
Methods
Datasets
Sources
Light
Dark
DRTune
anchor
Findings
IC-645
CLIP, PickScore, and HPSv2 text embeddings share a common direction (cone effect) that captures text-irrelevant preferences, and the orthogonal component c⊥p better measures T2I alignment; CLIP's untrained common direction makes it ineffective for reward fine-tuning
[supporting]