Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Watching, Reasoning, and Searching: A Video Deep Research Benchmark on Open Web for Agentic Video Reasoning

Chengwen Liu, Xiaomin Yu, Zhuoyue Chang, Zhe Huang, Shuo Zhang, Heng Lian, Kunyi Wang, Rui Xu, Sen Hu, Jianheng Hou, Hao Peng, Chengwei Qin, Xiaobin Hu, Hong Peng, Ronghao Chen, Huacan Wang

TL;DR

VideoDR introduces the first open-domain benchmark that anchors video reasoning to open-web evidence, requiring cross-frame cue extraction, iterative web retrieval, and multi-hop reasoning to produce verifiable answers. The framework compares Workflow and Agentic paradigms across six semantic domains, long and short videos, and multiple models, using an LLM judge to assess correctness. Key findings show that Agentic is not universally superior; success hinges on maintaining initial video anchors and managing goal drift over long retrieval chains, with long videos amplifying these challenges. The work outlines how current models struggle with long-horizon consistency and provides a practical benchmark to drive next-generation video deep research agents toward more stable, evidence-grounded reasoning on the open web.

Abstract

In real-world video question answering scenarios, videos often provide only localized visual cues, while verifiable answers are distributed across the open web; models therefore need to jointly perform cross-frame clue extraction, iterative retrieval, and multi-hop reasoning-based verification. To bridge this gap, we construct the first video deep research benchmark, VideoDR. VideoDR centers on video-conditioned open-domain video question answering, requiring cross-frame visual anchor extraction, interactive web retrieval, and multi-hop reasoning over joint video-web evidence; through rigorous human annotation and quality control, we obtain high-quality video deep research samples spanning six semantic domains. We evaluate multiple closed-source and open-source multimodal large language models under both the Workflow and Agentic paradigms, and the results show that Agentic is not consistently superior to Workflow: its gains depend on a model's ability to maintain the initial video anchors over long retrieval chains. Further analysis indicates that goal drift and long-horizon consistency are the core bottlenecks. In sum, VideoDR provides a systematic benchmark for studying video agents in open-web settings and reveals the key challenges for next-generation video deep research agents.

Watching, Reasoning, and Searching: A Video Deep Research Benchmark on Open Web for Agentic Video Reasoning

TL;DR

VideoDR introduces the first open-domain benchmark that anchors video reasoning to open-web evidence, requiring cross-frame cue extraction, iterative web retrieval, and multi-hop reasoning to produce verifiable answers. The framework compares Workflow and Agentic paradigms across six semantic domains, long and short videos, and multiple models, using an LLM judge to assess correctness. Key findings show that Agentic is not universally superior; success hinges on maintaining initial video anchors and managing goal drift over long retrieval chains, with long videos amplifying these challenges. The work outlines how current models struggle with long-horizon consistency and provides a practical benchmark to drive next-generation video deep research agents toward more stable, evidence-grounded reasoning on the open web.

Abstract

In real-world video question answering scenarios, videos often provide only localized visual cues, while verifiable answers are distributed across the open web; models therefore need to jointly perform cross-frame clue extraction, iterative retrieval, and multi-hop reasoning-based verification. To bridge this gap, we construct the first video deep research benchmark, VideoDR. VideoDR centers on video-conditioned open-domain video question answering, requiring cross-frame visual anchor extraction, interactive web retrieval, and multi-hop reasoning over joint video-web evidence; through rigorous human annotation and quality control, we obtain high-quality video deep research samples spanning six semantic domains. We evaluate multiple closed-source and open-source multimodal large language models under both the Workflow and Agentic paradigms, and the results show that Agentic is not consistently superior to Workflow: its gains depend on a model's ability to maintain the initial video anchors over long retrieval chains. Further analysis indicates that goal drift and long-horizon consistency are the core bottlenecks. In sum, VideoDR provides a systematic benchmark for studying video agents in open-web settings and reveals the key challenges for next-generation video deep research agents.
Paper Structure (36 sections, 3 equations, 4 figures, 5 tables)

This paper contains 36 sections, 3 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Our main contributions are as follows:
  3. Related Work
  4. Deep Research Benchmarks.
  5. Video Reasoning Benchmarks.
  6. Video Deep Research
  7. Task Definition
  8. Data Annotation Process
  9. Candidate Video Pool.
  10. Initial Filtering.
  11. Question Design.
  12. Quality Control
  13. Video & Web Dependency Testing
  14. Web-only Test.
  15. Video-only Test.
  16. ...and 21 more sections

Figures (4)

  • Figure 1: Overview of the VideoDR construction pipeline.
  • Figure 2: An example of the VideoDR task: identifying a museum via video visual cues, then using multi-hop search to find the closest "don't miss" exhibit to the entrance and outputting its accession number WB.67.
  • Figure 3: Human solvability across benchmark difficulty levels.
  • Figure 4: Data statistics of VideoDR, including (a) video category, (b) question length, and (c)video duration.