Bongard-OpenWorld: Few-Shot Reasoning Benchmark for Real-World Visual Concepts
The author introduces Bongard-OpenWorld as a benchmark to evaluate real-world few-shot reasoning for machine vision, emphasizing the challenges posed by open vocabulary and complex visual concepts.