Vision-language models fail at counting beyond training data

Original: Unveiling the Visual Counting Bottleneck in Vision-Language Models

Source: arxiv.org ↗