← back to paper
arxiv: 2510.20477 · 2 revisions
Bi-CoG: Bi-Consistency-Guided Self-Training for Vision-Language Models