Data Diversity Matters More Than Data Quantity in AI

In order to verify the effect of generating data variety in instance segmentation, this part tests DiverGen on the LVIS dataset. Experiments show that improving data diversity—through category, prompt, and model variation—drives sustained accuracy improvements, but increasing data quantity alone eventually plateaus or lowers performance.