COUNT ANYTHING

Summary from the TLDR AI newsletter

Object counting is still fragmented across domain-specific data sets and task formulations. Existing counting models are often tailored to specific scenarios and struggle to generalize across categories, visual domains, object scales, and density distributions. This paper presents a generalist model for text-guided object counting that achieves strong accuracy and multi-domain generalization.