OmniCompass

OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding

Xuyang Liu1,2Shihao Wang1Xinyu Wei1,2Minghan Li3Zeqing Wang1,2Zuhao Yang4Zichen Wen5Lei Zhang1,2
1 The Hong Kong Polytechnic University 2 OPPO Research Institute 3 Colorado School of Mines 4 Nanyang Technological University 5 Shanghai Jiao Tong University
The Hong Kong Polytechnic University Colorado School of Mines Shanghai Jiao Tong University

Overview

One benchmark, four dimensions

Benchmark Comparison

Table 1

Leaderboard

50 models · 3 tasks

From video selection to verified annotations

A closer look at the data

Examples

Videos, questions, predictions

QA Types

Model predictions across the two tracks

Analysis

What changes model performance?

Citation

Preprint · 2026