FNI V2.0 for Mavos Dd: Authority (A:62), Popularity (P:54), Recency (R:88), Quality (Q:50). Semantic (S) is a query-time baseline scored live at search.
LICENSE: This dataset is released under the CC BY-NC-SA 4.0 license.
This repository contains MAVOS-DD an open-set benchmark for multilingual audio-video deepfake detection.
Below, you can find the code to obtain the subsets described in the paper: train, validation, open-set model, open-set language and open-set full:
{python}
from datasets import Dataset, concatenate_datasets
metadata = Dataset.load_from_disk('MAVOS-DD')
metadata_indomain = metadata.filter(lambda sample: sample['split']=='test' and not sample['open_set_model'] and not sample['open_set_language'])
metadata_open_model = metadata.filter(lambda sample: sample['split']=='test' and sample['open_set_model'] and not sample['open_set_language'])
metadata_open_model = concatenate_datasets([metadata_indomain, metadata_open_model])
metadata_open_language = metadata.filter(lambda sample: sample['split']=='test' and not sample['open_set_model'] and sample['open_set_language'])
metadata_open_model = concatenate_datasets([metadata_indomain, metadata_open_language])
metadata_all = metadata.filter(lambda sample: sample['split']=='test')
The scripts require the datasets package to be installed.
{bash}
pip install datasets
We provide two scripts: metadata_generation.py and dataset.py.
The metadata_generation.py script is responsible for generating the metadata. Below is a sample metadata entry: