You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

FBench

Video-QA evaluation tables. Config longvideobench has a validation split; videomme has a test split. The repository contains Parquet tables, not bundled video files.

from datasets import load_dataset

data = load_dataset(
    "Xnhyacinth/FBench", "longvideobench",
    split="validation", revision="939bbeb50d47d4e6005ea882aa31adb8fcdc3c1f",
)

Related resources: ResAdapt and training mixture.

License and attribution

Config longvideobench is derived from LongVideoBench, licensed CC-BY-NC-SA-4.0 (attribution required, non-commercial, share-alike).

Config videomme is the Video-MME test split. It is retained here for benchmark comparison only and remains subject to the Video-MME upstream terms; no separate licence is granted for it by this repository. The repository contains question tables and metadata, not bundled video files.

Downloads last month
58

Collection including Xnhyacinth/FBench