Zi-Yuan Hu, Shuo Liang, Duo Zheng +10
The paper introduces the Needle in a Montage (NeMo) task and the NeMoBench benchmark to evaluate temporal understanding in video-language models, using an automated pipeline to gen…