ResearchBenchmarksVision-Language Compositionality on WhatsUp B (1 × 4 groups)Follow49.94GroupScoreTTM30.117635.263840.4145.5562Oct 9, 2025Evaluation ResultsMethodMethodLinksGroupScoreAbsolute Gain (Δ)Relative Gain (%)Error Reduction (%)TTMBackbone=CLIP-B32, Alg...Backbone=CLIP-B32, Algorithm=Algorithm 12025.1049.9419.161.727.6CLIP-B32Backbone=CLIP-B32Backbone=CLIP-B322025.1030.88———