Daily-Omni: Towards Audio-Visual Reasoning with Temporal Alignment across Modalities
Paper • 2505.17862 • Published
Dataset Viewer
The dataset viewer is not available for this dataset.
Unexpected token '<', "<html>
<h"... is not valid JSON
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Unify-OmniBench
统一格式的多模态评测数据集，由 Unify-OmniBench 框架转换生成。 包含四个 benchmark，在 Dataset Viewer 右上角下拉框切换。
数据概览
| Config (bench) | 题目数 | 模态 | 媒体 |
|---|---|---|---|
daily_omni |
~1197 | Video + Audio | .mp4 + .wav |
omnibench |
~1142 | Image + Audio | .png/.jpg + .mp3 |
omnivideobench |
~1000 | Video (embedded audio) | .mp4 |
worldsense |
~3172 | Video (embedded audio) | .mp4 |
使用方式
from datasets import load_dataset
ds = load_dataset("yukiyounai/Unify-OmniBench", "daily_omni", split="train")
print(ds[0]["question"], "->", ds[0]["answer"])
引用
- OmniBench: arXiv 2409.15272
- Daily-Omni: arXiv 2505.17862
- OmniVideoBench: NJU-LINK/OmniVideoBench
- WorldSense: via VLMEvalKit
- 转换工具: Unify-OmniBench
- Downloads last month
- 1,903