Search:
Match:
1 results

Analysis

This paper addresses the limitations of current Vision-Language Models (VLMs) in utilizing fine-grained visual information and generalizing across domains. The proposed Bi-directional Perceptual Shaping (BiPS) method aims to improve VLM performance by shaping the model's perception through question-conditioned masked views. This approach is significant because it tackles the issue of VLMs relying on text-only shortcuts and promotes a more robust understanding of visual evidence. The paper's focus on out-of-domain generalization is also crucial for real-world applicability.
Reference

BiPS boosts Qwen2.5-VL-7B by 8.2% on average and shows strong out-of-domain generalization to unseen datasets and image types.