OmniVideo-Test: 505 Human-Verified Video QA Pairs for AI Evaluation

OmniVideo-Test is a human-verified evaluation set introduced in the paper 'OmniVideo-100K: A Dataset for Audio-Visual Reasoning through Structured Scripts and Evidence Chains'. It contains 505 multiple-choice question-answer pairs based on raw video files, serving as a companion to the larger OmniVideo-100K dataset. The dataset was created by MiG-NJU and was last updated on June 15, 2026.

Use Cases

• Benchmarking video question-answering models based on the 505 human-verified QA pairs.
• Evaluating multimodal reasoning capabilities using the provided task taxonomies and evidence chains.
• Training or fine-tuning models for structured script understanding from video content.

Strengths

• Includes 505 human-verified multiple-choice QA pairs, providing a clear evaluation target.
• Contains raw video files, offering direct access to the source multimedia data.
• Provides structured metadata such as task taxonomies, ground-truth answers, and options for each question.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

MiG-NJU, as indicated in the repository link.

Collection Method

Created as a human-verified test set for the OmniVideo-100K research paper.

Freshness

Last updated 2026-06-15 02:13:16; freshness should be verified.