🤖 AI Summary
This work addresses the common misuse of the original Cosine Similarity Distance (CSD) as an absolute measure of style fidelity, highlighting its lack of reliable calibration for artist-level style discrimination. The authors propose “discriminative gap”—a diagnostic method that requires neither prototypes nor thresholds—to evaluate CSD’s effectiveness in distinguishing intra- versus inter-style pairs. For the first time, they reveal shared limitations across mainstream vision backbones (CLIP, SigLIP, DINOv2) in style discrimination and introduce the CSD+ protocol, which integrates CSLS reweighting, positional embedding interpolation to 336 pixels, and a frozen backbone strategy. Experiments across 91 artists show that while the original CSD yields negative discriminative gaps for 15 artists, CSD+ reduces this to only 4 and improves unsupervised pairwise verification AUC from 0.883 to 0.905.
📝 Abstract
Raw cosine in the 768-dimensional output space of the Contrastive Style Descriptor (CSD) is now widely read as an absolute, calibrated style-fidelity score for text-to-image and style-imitation evaluation. We introduce the discrimination gap, a corpus-internal, prototype-free and threshold-free diagnostic that tests whether contrastive style cosines admit an absolute same-versus-different interpretation on a candidate artist corpus. On a 1799-artwork, 91-artist public-domain corpus, raw CSD cosine yields negative point-estimate gaps for $23/91$ artists at the pairwise level ($2/91$ robust under bootstrap) and for $15/91$ in the aggregated-pool scoring regime style-fidelity evaluations typically use. CSLS readout on the frozen backbone reduces the aggregated negative-gap count to $4/91$; combined with positional-embedding interpolation to $336$ pixels it raises unsupervised pair-verification AUC from $0.883$ to $0.905$ across $25$ artist-disjoint splits. We refer to this diagnostic-driven readout protocol on the frozen backbone (CSLS as default, pos-interp $336$ as the stronger optional setting) as CSD+, not a new encoder.A cross-backbone check on CLIP-ViT-L/14, SigLIP-large and DINOv2-Large reproduces the same shared-tradition failure pattern, providing evidence that the residual reflects a shared limitation of the four backbones we tested rather than a CSD-specific artefact. Practical implication: before reporting CSD cosine as an absolute style-fidelity score, run the diagnostic on the candidate corpus; CSLS is the minimal correction when it fails.