作者位次 #5/8ACL 2026Findings· Findings of ACL
F ine S tate-Bench: Benchmarking State-Conditioned Grounding for Fine-grained GUI State Setting
Fengxian Ji · Jingpu Yang · Zirui Song · Yuanxi Wang · Zhexuan Cui · Yuke Li · Qian Jiang · Xiuying Chen
摘要
Despite the rapid progress of large vision-language models (LVLMs), fine-grained, state-conditioned GUI interaction remains challenging. Current evaluations offer limited coverage, imprecise target-state definitions, and an overreliance on final-task success, obscuring where and why agents fail.T...