Evolved agent harnesses cut long-video analysis costs while improving accuracy

VidHarness automatically searches for video-agent programs that select informative frames under different computation budgets.

Chinese Tech
Susan Liang · Jianmin Wu · Daxiang Dong

Baidu, Inc.

Research Digest··2 min read
Liang, Wu and Dong automate the design of agent harnesses, programs that control which video frames a frozen vision-language model observes and when it answers.

The authors built VidHarness around a proposer that writes candidate harnesses, runs them under a strictly enforced frame budget, examines execution feedback and revises the programs.

Why this paper

From Baidu, Inc. · Part of Agent Harness Optimization, now 90 papers

In one line

VidHarness automatically evolves video agent harnesses, achieving state-of-the-art long video QA with fewer frames.

What we could check

  • ·No code link found
  • ·No weights link found
  • ·No dataset link found
  • ·No compute details found
  • ·No stated limitations found
  • ✓Reports numbers on named benchmarks (3 benchmarks)

Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.

§
newspaper

Research Digest

Articles published under the Zotpaper byline are synthesized from multiple source publications by our AI editor and reviewed by our editorial process. Each story combines reporting from credible outlets to give readers a balanced, comprehensive view.