arxivcs.CV2026-07-21
ExpertVerse: A General-Purpose Benchmark for Expert-Level Reasoning in Knowledge-Intensive Visual Synthesis
Yuan Wang, Yongchao Du, Mengting Chen, Jinsong Lan, Xuetao Feng, Xiaoyong Zhu
Recent advances in multimodal generative models have enabled instruction-based image generation to move beyond semantic manipulation to knowledge-driven visual reasoning. However, these methods focus on explicit commonsense reasoning, shallow causal understanding, and direct know…