Datasets:
You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
NeevVision-v1 contains face crops derived from FaceForensics++, Google/Jigsaw DeepFakeDetection, DF40 and FFHQ, showing real people. Access is approved manually, only for researchers who are approved for FaceForensics++ and DF40 by their authors and accept their terms of use. No redistribution.
Log in or Sign Up to review the conditions and access this dataset content.
NeevVision-v1
Face crops for deepfake detection (binary real/fake + 6-class), merged from three datasets with one schema. 811,652 images. Access is approved manually; the source datasets' terms of use apply (FaceForensics++, DF40: research only, no redistribution; FFHQ: CC BY-NC-SA 4.0).
| column | values |
|---|---|
image |
face crop (PNG bytes copied unchanged from the sources) |
binary_label |
0 real, 1 fake |
class_id |
0 real, 1 FS, 2 FR, 3 EFS, 4 FE, 5 PA |
binary_name |
real / fake |
class_name |
real, FS, FR, EFS, FE, PA |
Sources and splits
| name | repo | split |
|---|---|---|
| ffpp | keyur-neeved/ff-filtered @ 4b06ca781f |
kept (official FF++ split; DFD by actor) |
| df40 | keyur-neeved/df40-faces @ b088b7d215 |
split here 70/15/15 per class |
| ffhq | arjunsingh-neeved/FFHQ-Dataset-FaceCrops @ e469988b24 |
split here 70/15/15 per class |
- Split per dataset and per class; DF40 frames of one video (folder) stay in one split (folders with
512 images are split per image); DF40 videos named with FF++ ids follow the official FF++ split, like
ff-filtered. - Rows are shuffled across datasets and classes inside every shard (seed 2026).
- Provenance of every row (source repo, file, row):
metadata/groups/*.parquet; split plan:metadata/split_plan.parquet.
Rows per class
| class | train | validation | test | total |
|---|---|---|---|---|
| real | 96,419 | 20,209 | 20,192 | 136,820 |
| FS | 206,609 | 42,416 | 43,498 | 292,523 |
| FR | 158,237 | 33,214 | 32,981 | 224,432 |
| EFS | 110,514 | 23,682 | 23,681 | 157,877 |
| FE | 0 | 0 | 0 | 0 |
| PA | 0 | 0 | 0 | 0 |
Rows per dataset and class
| dataset / class | train | validation | test | total |
|---|---|---|---|---|
| df40 / EFS | 110,514 | 23,682 | 23,681 | 157,877 |
| df40 / FR | 116,045 | 24,867 | 24,867 | 165,779 |
| df40 / FS | 84,462 | 18,099 | 18,099 | 120,660 |
| df40 / real | 31,791 | 6,813 | 6,812 | 45,416 |
| ffhq / real | 37,108 | 7,952 | 7,952 | 53,012 |
| ffpp / FR | 42,192 | 8,347 | 8,114 | 58,653 |
| ffpp / FS | 122,147 | 24,317 | 25,399 | 171,863 |
| ffpp / real | 27,520 | 5,444 | 5,428 | 38,392 |
Usage
from datasets import load_dataset
ds = load_dataset("keyur-neeved/NeevVision-v1", token=True) # or streaming=True
x = ds["train"][0]
x["image"], x["binary_label"], x["class_id"]
- Downloads last month
- 68