Count Anything

Researchers introduced Count Anything, a text-guided object counting model designed for cross-domain generalization.

They also released CLOC, a benchmark assembled from diverse public datasets spanning six domains.

The model combines sparse region-level and dense pixel-level counting with complementary fusion to better handle different object sizes and densities.

This advances open-world counting and could support applications in remote sensing, histopathology, agriculture, and microscopy.