MultiView-Bench: A Diagnostic Benchmark for World-Centric Multi-View Integration in VLMs

Hantao Zhang, Jinru Sui, Ed Li +2

Recent benchmarks for VLMs largely assess single- or limited-view perception, leaving untested the core cognitive ability to integrate observations across viewpoints into a coheren…