Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: From Evaluation to Diagnosis

Recent advancements in Large Vision-Language Models (LVLMs) have demonstrated remarkable multimodal perception and reasoning capabilities. While numerous benchmarks have evaluated LVLMs from holistic or task-specific perspectives, their capabilities on fine-grained image tasks-fundamental to compute...