New AI model called "Count Anything" does exactly what it says, and that's harder than it sounds

The "Count Anything" AI model has been released, designed to count objects in various image types solely based on text prompts. It achieves a significant reduction in error rates, halving the performance issues seen in prior models, though it still faces challenges with densely packed objects and ambiguous terminology. This advancement is crucial for practitioners as it broadens the applicability of AI in image analysis across diverse fields, from crowd counting to microscopic evaluations.