keyur-neeved/df40-filtered

DF40 — schema preparation only

Status: COMPLETE — all expected rows verified and uploaded.

Source: kd13/df40-unfiltered. Pinned commit: 545ef986f529a1306476651a82d316d18f72d7a9. Source configuration: images.

All rows from train, unassigned, and multi_split_reference are merged into configuration image, split train. Exactly five columns: image, binary_label, class_id, binary_name, class_name.

Original image bytes, image paths, resolutions, row multiplicity, and within-file order are preserved. No cleaning, quality filtering, deduplication, image decoding/re-encoding, cropping, resizing, balancing, augmentation, or shuffling is performed. The old class ID is replaced by the mapping below; binary labels are converted to int64 and checked against each row's family. Unknown or inconsistent labels stop the conversion.

class_id	class_name	binary_label	binary_name
0	real	0	real
1	FS	1	fake
2	FR	1	fake
3	EFS	1	fake
4	FE	1	fake
5	PA	1	fake

FS = face swapping; FR = face reenactment; EFS = entire face synthesis; FE = face editing; PA = presentation attack. No rows are fabricated for absent classes.

_prepare/state.json records the pinned source file, source split, counts, and integrity checks for every output file. Output file row i corresponds to source file row i. These small support files are not additional dataset subsets.

The merged split name is a storage choice, not proof of an independent evaluation split. Original duplicates and unresolved split memberships remain. No Indian identity or single-person eligibility is established by this transformation.

Source provenance and terms: original source card and DF40 project. No new license is assigned. Only the source repository's image configuration is copied; this is not a new claim that every original DF40 release component is present.

Verified row counts

Original split	Rows
train	604,772
unassigned	27,031
multi_split_reference	10,099
Merged train	641,902

class_id	class_name	Rows
0	real	60,123
1	FS	163,518
2	FR	223,960
3	EFS	194,301
4	FE	0
5	PA	0

Load the dataset

from datasets import load_dataset
ds = load_dataset('keyur-neeved/df40-filtered', name='image', split='train', token=True)

Use streaming=True to avoid downloading the complete dataset. For a private repository, authenticate in your own environment or pass your token securely.