Skip to content

Dataset · JoeLeelyf

Streaming video understanding benchmark for Video-LLMs

JoeLeelyf/OVO-Bench

A benchmark evaluating video models on real-time video question answering (CVPR 2025).

What's inside

The benchmark evaluates online video understanding across three modes, including backward tracing of past events and real-time visual perception. Contains 1 to 10 thousand English entries. Accepted at CVPR 2025, with the current repository version updated compared to the original arXiv preprint.

What you can do with it

  • Evaluating Video-LLMs on streaming video analysis and past-frame memory tasks
  • Comparing multimodal models on a current academic benchmark
Tasks
Video text → text
Languages
English
Size
1K–10K rows
Downloads this month
23.4K
Likes
13

How to load it

from datasets import load_dataset

ds = load_dataset("JoeLeelyf/OVO-Bench")