Feature
Researchers Propose Explicit Granularity for VLM Object Counting
Amy Reed · 12 May 2026
Vision-language models struggle with reliable open-world object counting despite rapid advances. Researchers attribute brittleness to implicit granularity and propose explicit user specification.
Researchers released a paper on open-world object counting, noting that vision-language models remain brittle due to implicit granularity in user references. Users might mean specific identities, attributes, or other levels, leading to unreliable counts. The work pushes for explicit handling of any counting…