LeeMirror013/GravureAtlas

Gravure Atlas

Gravure Atlas is an image archive and catalog organized by album and source. It contains the retrieved image files, a Parquet index for each shard, and the SQLite catalog with album, source, and image-location records. Image files keep the bytes returned by their source URLs; the export does not resize or re-encode them.

The catalog contains 5,712,029 image rows. The published image set covers 5,408,933 eligible rows from 41,355 albums. It contains 5,076,648 image files. The other 332,285 eligible rows remain in the Parquet indexes with status=failed and no image member.

Files

Path	Contents
images/0000.tar through images/0999.tar	Retrieved image files, grouped into 1,000 shards. Each member is named <image_id>.<extension>.
metadata/0000.parquet through metadata/0999.parquet	One row per eligible image candidate, including candidates whose image could not be retrieved or validated.
photos.sqlite3	The SQLite catalog, including source rows excluded from the published image set.

An image belongs to shard image_id % 1000. The Parquet id is the numeric image ID used in tar member names and shard assignment. In the SQLite catalog, image rows are keyed by (source_id, position); the numeric ID follows that row order.

Metadata

The Parquet files include the image ID, source_id, position, album_id, album title and descriptive fields, source site and URLs, retrieval status, attempt count, and HTTP or validation error details. Successful rows also include tar_path, member_name, image_bytes, width, and height. Join a successful metadata row to its tar member using member_name; failed rows have no tar member.

The SQLite database has three tables:

Table	Fields
images	source_id, position, url; the primary key is (source_id, position).
album_sources	id, album_id, site, source_url.
albums	id, title, model_name, series, issue.

Selection

The image shards and Parquet indexes omit albums whose title contains the explicit label AI Generated, allowing spaces, underscores, or hyphens and ignoring case. This rule excludes 437 albums and 303,096 image rows. Labels such as AI Enhanced and AI restored do not match the rule. The raw photos.sqlite3 retains the excluded albums and their image rows.

Access

Download a shard and its index with huggingface_hub:

from huggingface_hub import hf_hub_download
import pyarrow.parquet as pq

repo_id = "LeeMirror013/GravureAtlas"
tar_path = hf_hub_download(repo_id, "images/0000.tar", repo_type="dataset")
index_path = hf_hub_download(repo_id, "metadata/0000.parquet", repo_type="dataset")
metadata = pq.read_table(index_path)

Image source URLs and album details refer to the catalog's source sites. Rights to the images remain with their respective owners; check applicable permissions and source-site terms before reuse.