bvd-urls-sample / README.md
flashdata-dev's picture
Fix license and checksum file links
7ce5a6e verified
|
Raw History Blame Contribute Delete
5.77 kB
metadata
license: cc-by-4.0
size_categories:
  - n<1K
pretty_name: BVD-URLs — FlashData YouTube Reference Sample
tags:
  - video-metadata
  - youtube
  - flashdata
  - custom-video-dataset
configs:
  - config_name: default
    data_files:
      - split: sample
        path: sample.parquet

BVD-URLs — FlashData YouTube Reference Sample

256 YouTube source references with Common Crawl provenance for testing URL normalization, deduplication and dataset intake.

256 records · URL and crawl metadata · Parquet + JSONL · No video files

Request a custom video dataset · Explore the YouTube Download API

Use this sample to inspect a source manifest, test a loader and prepare a scoped collection request. For playable MP4 examples, see FlashData's video dataset Collection.

Preview

Video ID Example metadata
JFUsP1coIKM CC-MAIN-20200122071919-20200122100919-00278.warc.gz
Cs1Dm6ELYjQ CC-MAIN-20190824194521-20190824220521-00077.warc.gz
p1AwYoA8Yfo CC-MAIN-20190319183735-20190319205735-00104.warc.gz
Y32cDIwFt4k CC-MAIN-20191018104351-20191018131851-00134.warc.gz

Load the sample

from datasets import load_dataset

sample = load_dataset("flashdata-dev/bvd-urls-sample", split="sample")
print(len(sample))  # 256
print(sample[0]["source_video_url"])

Install datasets first if needed. Direct files: Parquet, JSONL, provenance, checksums, quality report.

Prepare your source list

Download prepare_delivery.py and sample.jsonl into the same folder, then run:

python prepare_delivery.py sample.jsonl --output delivery-manifest.jsonl --report normalization-report.json

The helper accepts local CSV or JSONL, normalizes YouTube IDs, deduplicates sources and accounts for every input row. It makes no network requests and does not download media. The output is a source manifest for scoping a delivery, not an API request payload.

For a custom collection, provide the source list or topic, target languages, desired resolution, clip boundaries, caption needs and storage destination. FlashData can scope available media delivery and metadata packaging against the requested permissions and acceptance criteria. See the YouTube Download API documentation for the current API contract.

Fields

Field Meaning
video_id YouTube ID parsed from the source URL
source_video_url Canonical YouTube watch URL
uid, url Unmodified upstream record ID and URL
cc_filename, domain Unmodified upstream crawl filename and platform domain
source_row_index Zero-based row in the pinned upstream shard

Source and selection

Upstream: https://huggingface.co/datasets/laion/BVD-URLs. Exact source revision and record positions are in provenance.json.

First 256 distinct valid YouTube IDs encountered in the first 2,048 records of default/part-000.parquet. Input order is retained; this is not a random or representative sample.

Changes made by FlashData: Retained uid, url, cc_filename and domain. Omitted alt and page_url. Added video_id, canonical source_video_url and zero-based source_row_index. Canonical URLs remove embed-player query parameters; the original url is preserved.

License and attribution

The public laion/BVD-URLs repository labels this URL/metadata release CC BY 4.0. This package follows that scope. LAION's separately gated video releases and protected research environment have additional research-only conditions; those media are not part of this package. A crawl record does not establish current availability or a source video's reuse license.

LAION-BVD authors: Andreas Hochlehnert, Marianna Nezhurina, Mehdi Cherti, Andrej Radonjic, Thaddäus Wiedemer, Christoph Schuhmann, Romain Beaumont, Wieland Brendel, Bernhard Schölkopf, A. Sophia Koepke, Jenia Jitsev and Matthias Bethge. URL collection derived from Common Crawl.

See LICENSE, ATTRIBUTION.md and LICENSE-CODE. FlashData selected and packaged this sample independently. No affiliation or endorsement by the upstream projects or creators is implied.

Validation and limits

All 256 records load locally through Hugging Face Datasets; JSONL and Parquet match exactly. IDs are unique and syntactically valid. Every package file has a SHA-256 checksum. No source video was downloaded or checked for current availability, and no transcript or annotation accuracy score is claimed. Upstream content, availability and rights may differ across records. This small sample is intended for inspecting formats and workflows.

References