VideoKR: 315K Knowledge- and Reasoning-Intensive Video Examples

VideoKR-Train is a large-scale training corpus for knowledge- and reasoning-intensive video understanding. It contains 315,000 video reasoning examples derived from 145,000 newly collected, CC-licensed videos. The dataset was created by minuzero and presented at ICML 2026.

Use Cases

• Train video-language models for complex reasoning based on the described knowledge-intensive examples.
• Benchmark model performance on video-based question answering and inference tasks.
• Develop and evaluate multimodal AI systems that integrate visual and world knowledge.

Strengths

• 315,000 video reasoning examples provide a substantial scale for training.
• Specifically designed for knowledge- and reasoning-intensive tasks, a stated focus area.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Newly collected, CC-licensed videos.

Freshness

Last updated 2026-06-05 01:43:00; freshness should be verified.