First, we propose Visual Feature Guidance with CLIP, which enriches and enhances the global representation capabilities of backbone while preserving its self-learning ability for visual ...