You can deploy this model and start earning money today!

SOBER137/
Video-Holmes

Views

No views yet

Video-Holmes: Can MLLM Think Like Holmes for Complex Video Reasoning?

Junhao Cheng1,2, Yuying Ge1,✉, Teng Wang1,✉, Yixiao Ge1, Jing Liao2, Ying Shan1

1ARC Lab, Tencent PCG, 2City University of Hong Kong

🔎 Introduction

Video-Holmes is a benchmark designed to evaluate the complex video reasoning capabilities of MLLMs. Video-Holmes consists of 1,837 questions derived from 270 manually annotated suspense short films (ranging from 1… See the full description on the dataset page: https://huggingface.co/datasets/SOBER137/Video-Holmes.