JamalLee/pre-2026

release-prep-2026

A large-scale image dataset for image classification and forgery/manipulation detection research, packaged as Parquet shards.

Structure

Data is organised under data/ into modality subfolders:

• data/Image/ - image-modality Parquet shards ({split}-XXXXX-of-YYYYY.parquet).
• data/Video/, data/Audio/, data/AV-TH/ - reserved for future modalities.

field	type	description
image	image	The input image.
mask	image	Optional binary mask for manipulated regions (may be null).
label	string	Category: real, full_synthetic, tampered, or mask.
generator	string	Source / generator model (null for real images).
filename	string	Relative path of the original image within its source split.
split	string	One of train, validation, test, masks.

Splits

• train, validation: the main labelled set.
• test: out-of-distribution evaluation images.
• masks: auxiliary manipulation masks.

Usage

from datasets import load_dataset
ds = load_dataset("JamalLee/release-prep-2026")
print(ds)

License

Released under the Creative Commons Attribution 4.0 International License.