metadata
pretty_name: Keyframe Selection Benchmark
language:
- en
task_categories:
- video-classification
- visual-question-answering
Keyframe Selection Benchmark
This portable subset contains 10 TVSum videos, 15 SumMe videos, 13 EgoSchema clips with public offline answers, and the two shortest verified HourVideo development videos with all mirrored questions and answers. TVSum and SumMe retain their human summary annotations. The long form datasets retain their multiple choice annotations for downstream coverage evaluation.
manifest.jsonl is the normalized entry point. Paths are relative to the dataset root, and each record carries the annotation fields needed by its source benchmark.