submission11898/MockWeb-Bench-Media

MockWeb-Bench Media

Individually downloadable images and videos from MockWeb-Bench, version v3_relicensed_260825-assets.

Every url in the JSONL files is a public HTTPS download link. No token, login, or periodic signature renewal is required. Download clients should follow HTTP redirects; curl -L and Python requests support this.

Modality	Source records	Unique files	Unique bytes
Images	75,437	62,668	48,060,147,019
Videos	612	28	45,211,539
Total	76,049	62,696	48,105,358,558

Images cover all 50 source sites. Videos occur in 19 sites. Image records include website assets, full-page desktop/mobile screenshots, region screenshots, SVG, and icon files. Source records sharing a stored file reuse the same URL while retaining each original source path.

Download the manifests

• image.jsonl: one record per original image file.
• video.jsonl: one record per original video file.
• unique-files.jsonl: one record per stored media file, suitable for downloading the collection once without duplicates.
• mockweb-media-jsonl.zip: all three JSONL files, the manifest, this README, and source license/notice files.
• manifest.json: record counts, file sizes, SHA-256 hashes, and source versions.

JSONL fields

• url: public HTTPS download URL for the original file bytes.
• repo_path: file location in this repository.
• modality, mime_type: the source modality and content type.
• domain, path: the source site and original relative path, included in image/video records.
• size, sha256: original file length and SHA-256 digest.
• image_role: image purpose, included in image records.

Media URLs are pinned to commit c80510ff55cf80684d08e7771fbb54a541dd9926. They contain no credentials or expiration timestamp. Source paths and file contents are preserved. The source includes 7 zero-byte image records representing two unique files; these are preserved with size: 0.

For example, download the first video with Python:

import json
import requests

manifest_url = "https://huggingface.co/datasets/submission11898/MockWeb-Bench-Media/resolve/main/video.jsonl?download=true"
response = requests.get(manifest_url, timeout=60)
response.raise_for_status()
record = json.loads(response.text.splitlines()[0])
with requests.get(record["url"], stream=True, timeout=60) as media:
    media.raise_for_status()
    with open("video.mp4", "wb") as output:
        for chunk in media.iter_content(1024 * 1024):
            output.write(chunk)

Source and licensing

This release mirrors only the image and video files of source dataset commit b24902a1ff2d7333c8a10b9b8f00415ab5b9ea40. Website code, task definitions, evaluation metadata, and complete per-site archives remain in the source dataset.

The source per-site LICENSE/NOTICE files and license index are preserved under LICENSES. Licensing varies by site and component; the corresponding source notices apply.

All 62,696 stored files were checked against the original byte sizes and SHA-256 hashes before upload. Remote file sizes and content identifiers were then checked against the full inventory.