Long Videos for Omni-Modal Understanding Tasks

LongVALE contains 8,411 long videos totaling 549 hours for fine-grained video understanding. The dataset was created by ttgeng233 for academic research and was last updated in August 2025. It integrates vision, audio, language, and event data.

Use Cases

• Train models for video event recognition using the integrated event data.
• Evaluate audio-visual synchronization models using the aligned audio and video streams.
• Develop captioning systems using the associated language data from video descriptions.
• Benchmark long-form video understanding models on the 8,411 video samples.

Strengths

• 8,411 long video samples provide substantial volume for training.
• 549 total hours of video content offers extensive temporal coverage.
• Integrates four data modalities: vision, audio, language, and event.

Limitations

• Specific column names, file formats, and dataset size details are unknown.
• The dataset's geographic and temporal coverage for its video sources is not specified.
• Potential class imbalance or label noise cannot be assessed without sample data.

Provenance

Source

null

Collection Method

null

Time Range

null

Freshness

null

Geography

null