OmniPro: A Comprehensive Benchmark for Omni-Proactive Streaming Video Understanding

Ruixiang Zhao, Jie Yang, Zijie Xin, Tianyi Wang, Fengyun Rao, Jing LYU, Xirong Li

arXiv:2605.18577 · 2026-05-17 공개 · arXiv · PDF

video-understanding audio-visual multimodal-benchmark streaming-video non-speech-audio long-horizon-robustness omni-proactive modality-isolation

Abstract

Omni-proactive streaming video understanding, i.e., autonomously deciding when to speak and what to say from continuous audio-visual streams, is an emerging capability of omni-modal large language models. Existing benchmarks fall short in three key aspects: they rely primarily on visual signals, adopt polling or fixed-timestamp protocols instead of true proactive evaluation, and cover only a limited range of tasks, preventing reliable assessment and differentiation of omni-proactive streaming models. We present OmniPro, the first benchmark to jointly evaluate omni-modal perception, proactive responding, and diverse video understanding tasks. It comprises 2,700 human-verified samples spanning 9 sub-tasks and 3 cognitive levels, covering 6 basic video understanding capabilities. Notably, 84% of samples require audio signals (speech or non-speech), and each sample is annotated with modality-isolation labels to enable fine-grained multimodal analysis. We further introduce a dual-mode evaluation protocol: Probe mode assesses content understanding by querying the model before and after each ground-truth trigger, while Online mode evaluates full proactive ability by requiring models to autonomously decide when to respond in streaming input. Evaluating 11 representative models reveals three key findings: (1) audio provides consistent gains but with highly variable utilization across models, (2) performance degrades significantly over time, indicating limited long-horizon robustness, and (3) non-speech audio perception remains the weakest dimension.

한국어 요약

📋 한 줄 요약

**[Omni-Proactive Streaming Video / Benchmark]** OmniPro가 2,700 human-verified 샘플·9 sub-task·3 cognitive level의 첫 omni-proactive 벤치마크 — 84% 샘플이 audio 필요, Probe·Online dual-mode 평가, 11 모델 분석에서 audio·long-horizon·non-speech 등 발견.

🎯 핵심 기여도

💡 핵심 아이디어

Omni-proactive video understanding 평가의 발전은 audio 필수성·true proactive(자율 시점 결정)·다양 task 커버리지를 동시에 갖춰야 하며, Probe·Online dual-mode가 content 이해와 proactive 능력을 각각 분리 평가해 모델 차별화를 가능하게 한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Omni-proactive streaming의 첫 종합 벤치마크, dual-mode 평가의 분리 평가 원리, audio·long-horizon·non-speech 약점의 정량 진단으로 후속 연구 방향 제시. **한계**: 2,700 샘플의 도메인 커버리지, human verification 일관성, 11 모델의 시점적 한정.

🚀 실용적 활용