DiscoVL: Unveiling Disentangled C ross-Modal Representation Learning via Orthogonal Adversarial Regularization for V ision-Language Models

Benchmark Model Rank Results
prompt-engineering-on-imagenet-aDiscoVL–Top-1 accuracy %: 51.30
prompt-engineering-on-imagenet-rDiscoVL–Top-1 accuracy %: 78.00
prompt-engineering-on-imagenet-sDiscoVL–Top-1 accuracy %: 49.40
prompt-engineering-on-imagenet-v2DiscoVL–Top-1 accuracy %: 64.60