Multi-Task Long Video Understanding Benchmark with 11 Tasks

MLVU is a benchmark dataset for multi-task long video understanding, containing a test set with 11 different tasks. It was created by the MLVU organization and updated in September 2024.

Use Cases

• Evaluate model performance on the 11 different tasks within the MLVU-Test set.
• Benchmark long video understanding models on tasks like Sports Question Answering (SQA).
• Assess multi-task learning capabilities across diverse video understanding challenges.

Strengths

• Benchmark designed for 11 distinct video understanding tasks.
• Data released in 2024, indicating recent collection and relevance.

Limitations

• Dataset size, row count, and specific column structure are unknown.
• Limited to benchmark evaluation tasks, not raw video data for general training.

Provenance

Source

MLVU organization via Hugging Face.

Collection Method

Annotation data and evaluation code for a research paper benchmark.

Time Range

null

Freshness

Last updated 2024-09-18.

Geography

null