license: other
library_name: pytorch
tags:
- deepfake-detection
- 3d-avatar
- nerf
- 3d-gaussian-splatting
- gated-data
3DDF
This gated research release contains the code and the non-duplicate 3DDF data used for 3D-generated face forgery detection experiments. It is organized as independent archives so the Hub repository stays usable without committing more than a million individual files.
Included Data
- NeRF: SyncTalk, ERNeRF, RADNeRF
- 3DGS: InsTag, DEGSTalk, TalkingGaussian
- EG3D-family: Real3D, MimicTalk, EMOPortraits
- External evaluation: GAGAvatar, GPAvatar
- Real reference pool: HDTF/TFHP
- Extended methods: DynTet, Portrait4D-V1, Portrait4D-V2, GaussianHeadAvatar, Next3D
- Extended real pools: CelebHQ (local 7,000-image subset), NeRSemble, and FFHQ part-1 (local indices 00000-33669; 33,670 source images)
The strict release contains 233,863 cropped RGB frames, matching 81-point
landmark arrays, and the corresponding 5,146 source videos. Invalid or
unreferenced files are excluded using manifests/3ddf_strict.jsonl.
The non-duplicate extension adds 707,546 source files totaling 94,392,974,625 bytes (87.910 GiB) before archiving. It includes packaged processed frames/landmarks and the source images or videos needed to reproduce preprocessing.
The strict core is the immediately training-ready benchmark portion. The extension is a packaged asset collection, not a claim that every component is immediately training-ready. Its processed archives contain individual frames and matching landmark files, while its raw archives contain source images or videos for preprocessing. DynTet, Portrait4D-V1, and GaussianHeadAvatar include v6 JSON metadata; Portrait4D-V2, Next3D, and the real pools still require the appropriate split/config integration before direct use by the training entry point.
| Component | Processed files | Processed bytes | Raw files | Raw bytes | Subset note |
|---|---|---|---|---|---|
| DynTet | 4,348 | 204,987,710 B (0.191 GiB) | 68 | 109,613,449 B (0.102 GiB) | |
| Portrait4D-V1 | 136,938 | 7,708,739,093 B (7.179 GiB) | 70,000 | 5,153,354,994 B (4.799 GiB) | |
| Portrait4D-V2 | 138,890 | 7,384,174,579 B (6.877 GiB) | 70,000 | 4,454,085,516 B (4.148 GiB) | |
| GaussianHeadAvatar | 19,370 | 690,728,752 B (0.643 GiB) | 9,695 | 3,627,434,847 B (3.378 GiB) | |
| Next3D | 107,994 | 5,153,315,528 B (4.799 GiB) | 845 | 2,058,242,309 B (1.917 GiB) | |
| CelebHQ | 13,734 | 755,561,422 B (0.704 GiB) | 7,000 | 228,619,752 B (0.213 GiB) | Local subset of 7,000 source images. |
| NeRSemble | 18,554 | 912,746,627 B (0.850 GiB) | 9,648 | 5,955,395,153 B (5.546 GiB) | |
| FFHQ | 66,792 | 3,958,829,271 B (3.687 GiB) | 33,670 | 46,037,145,623 B (42.875 GiB) | Part-1 local subset, indices 00000-33669 (33,670 source images). |
| Total | 506,620 | 26,769,082,982 B (24.931 GiB) | 200,926 | 67,623,891,643 B (62.980 GiB) |
The extension deliberately excludes archive duplicates and alternate copies:
Portrait4D-V1-copy, Portrait4D-V2-copy, next3d_infer.zip, and
FFHQ-1024-1.zip. Portrait4D composite source copies are alternate
representations of the same logical samples and are not part of this
non-duplicate release.
Layout
code/DeepfakeBench/ cleaned source snapshot and configs
data/processed/ RGB frame + landmark archives by method
data/raw/ source video archives by method
data/extended/processed/ extended cropped frames + landmarks
data/extended/raw/ extended source images + videos
manifests/ relative-path manifests and split metadata
SHA256SUMS checksums for the complete release
Archives
| Archive | Size | SHA-256 |
|---|---|---|
data/extended/processed/CelebHQ-part-00000-of-00001.tar.zst |
0.696 GiB | a886a28d752984d163afd56398fa6ce6d42fef6881321803ededc6ae30701fe1 |
data/extended/processed/DynTet-part-00000-of-00001.tar.zst |
0.190 GiB | 0812c201d5e47563039371095ab1965ba722b7296560b44871e6ef7c521db442 |
data/extended/processed/FFHQ-part-00000-of-00001.tar.zst |
3.654 GiB | 21e3ba0ba670961284790be48e488c896b9ce91f345d8d771e851b2c4b7a3f6e |
data/extended/processed/GaussianHeadAvatar-part-00000-of-00001.tar.zst |
0.634 GiB | 9f3fabb704702553be171a292857de22ece36ba465e26b086df7888990658b2d |
data/extended/processed/NeRSemble-part-00000-of-00001.tar.zst |
0.847 GiB | c820f556bbec19a64aa5315b5df69d5da5e82ba70fca9925c133e583fb3ca006 |
data/extended/processed/Next3D-part-00000-of-00001.tar.zst |
4.773 GiB | f8c0aaacd825aa7e8c4a3d1e1edeeb2f72666a412afe076d3957979a81f71261 |
data/extended/processed/Portrait4D-V1-part-00000-of-00002.tar.zst |
6.496 GiB | e53ddb92866c15f5dd55f76fc49a8c17178cd4d87a7e8837c8a48000e2bc392e |
data/extended/processed/Portrait4D-V1-part-00001-of-00002.tar.zst |
0.680 GiB | 86562af3a1aa6fcefc32a47af70c5d33148d5438aaac09eb6a637b8489c0f03a |
data/extended/processed/Portrait4D-V2-part-00000-of-00002.tar.zst |
6.481 GiB | 1f6052d0e0d0351e6577ff2e8acddc7c07860755fab27d9e737337baad290f3b |
data/extended/processed/Portrait4D-V2-part-00001-of-00002.tar.zst |
0.376 GiB | 2fd5f2f2adff5040fd229f1bb313ca421543e5aad39e321851f66e83c5600147 |
data/extended/raw/CelebHQ-part-00000-of-00001.tar.zst |
0.209 GiB | 9812338653efdedb105413a19c0185aef956b3fa69735523bba743df2fdbd53d |
data/extended/raw/DynTet-part-00000-of-00001.tar.zst |
0.102 GiB | 47652c0426af17f1442499da89442070dbacaa41be74744a7d1c24a775a875ab |
data/extended/raw/FFHQ-part-00000-of-00007.tar.zst |
6.509 GiB | d2de181c9d9630506eeb8a3b9fc8f28dbe9510297749e1f24cbfa3bc94ce555b |
data/extended/raw/FFHQ-part-00001-of-00007.tar.zst |
6.508 GiB | 933ca0eadb7efcd4cfc6ccf85bf022a5b19e6a59f0fba181c02a28a54816768f |
data/extended/raw/FFHQ-part-00002-of-00007.tar.zst |
6.508 GiB | 7770491aece11798f634bf1eb94150f2ba323dae658785734906d981b58eaafd |
data/extended/raw/FFHQ-part-00003-of-00007.tar.zst |
6.508 GiB | c207702980a4cd34f3cb171ddea8e535ede2331d93ca7235e17c706bd2f9200b |
data/extended/raw/FFHQ-part-00004-of-00007.tar.zst |
6.508 GiB | e30ccff401dc26230c9bf48c1d47604a50a6725e4a87d309c89280119f366ee5 |
data/extended/raw/FFHQ-part-00005-of-00007.tar.zst |
6.507 GiB | b0c63b60c44ba4112084bef28c585abf5be78e658747014d40b0fd7d98b2025a |
data/extended/raw/FFHQ-part-00006-of-00007.tar.zst |
3.885 GiB | 9d2843eecb8374f4e5fd0db835caf19a3db5e8611a9040d868bf9711097540ff |
data/extended/raw/GaussianHeadAvatar-part-00000-of-00001.tar.zst |
3.099 GiB | edda82f42e4adaab48b1ad14f8cae742ae75c38776946f8ee0b740be8e1feaea |
data/extended/raw/NeRSemble-part-00000-of-00001.tar.zst |
5.448 GiB | 650ac793ff675c287425ab764b6f1b660ebc98094b417445a45678d61d1f94f6 |
data/extended/raw/Next3D-part-00000-of-00001.tar.zst |
1.914 GiB | a4f1333f42e45e843eff1f60b611d668b56d7f560ee620c5e547f56cebf69581 |
data/extended/raw/Portrait4D-V1-part-00000-of-00001.tar.zst |
4.695 GiB | cb396687a4cf42c405bee211cc55d680816b445b6c411c4b8801d42ed79fda42 |
data/extended/raw/Portrait4D-V2-part-00000-of-00001.tar.zst |
3.969 GiB | d03c3ac5256ee7b46ce89ec21f2c88874bcf1f6846ea7d02b8f04619e339d2eb |
data/processed/DEGSTalk.tar.zst |
1.146 GiB | f347a7019aebee38bc38c2f7357e8477b2c010a77e21f61018cc276ff00ad865 |
data/processed/EMOPortraits.tar.zst |
1.197 GiB | 8553abee263710bebdbd201643018661526a3fbe7f638610d4d3e131498ac8dc |
data/processed/ERNeRF.tar.zst |
2.413 GiB | 90e97a62fbfe4c1158c581b0fb4d029d515974dfcc96607dc8fdbb1510a59649 |
data/processed/GAGAvatar.tar.zst |
0.623 GiB | a0fa7abaa229fdaf09b83246b27c4e5d6374e48b2c1e3fe6a0bc980f755bfdc8 |
data/processed/GPAvatar.tar.zst |
0.872 GiB | aae2a205a971d863d4192a24fdfec10d6fa6e277e97357e0c8d0305233ef167d |
data/processed/InsTag.tar.zst |
1.054 GiB | 308c33c35e43de4bf40caaa8ac3cc29bef4f3978707b1d0774ad479acf1c4b94 |
data/processed/MimicTalk.tar.zst |
1.314 GiB | 7444bc72973a5cad4d6ad02eb8aff7dcb4d65e6cd030de9dd687770d9ea3b415 |
data/processed/RADNeRF.tar.zst |
1.095 GiB | 1438936ef7c091e5820c1ccf10e1dcee3ced30dd507df6b827b41278b48eecad |
data/processed/Real.tar.zst |
4.170 GiB | 3e89ebe179f127c4fcb012a867dd4f45b693950bd695d33d99549f10e4620304 |
data/processed/Real3D.tar.zst |
1.278 GiB | 551290edf2dea4f8877372da11fab02d4787615271b7c0c638f6f906e48a8667 |
data/processed/SyncTalk.tar.zst |
3.172 GiB | c71a30bec9f5bbb51d66276893c700989610b0dde541ea092ccae58493feff83 |
data/processed/TalkingGaussian.tar.zst |
1.058 GiB | dff252936e1596ed2e435f7597d3487bfc0d7001ff936214708ec2e397cc281f |
data/raw/DEGSTalk.tar.zst |
0.937 GiB | 4fd8c30ad1594e870b87d391c681157fad7ef383165d55370b4428eddce7e49e |
data/raw/EMOPortraits.tar.zst |
0.187 GiB | 1e3b6e4fc2ea628d38415cc0179cdb8dcf56f4a82e648a6b4ef716286a9f9dba |
data/raw/ERNeRF.tar.zst |
0.568 GiB | fdbc913e61931cf03e1da331b45b317990f5d6860e6638eedafbaa190aa9f191 |
data/raw/GAGAvatar.tar.zst |
0.369 GiB | 0ea325be064b403fe06b03a01382704c2a5ec1f31504ff1ca9dcbffa18bec6cf |
data/raw/GPAvatar.tar.zst |
0.225 GiB | 05e57f2b4b77bd8aec66e4dcd4778f5cda0aeff35481a7224c836e98113719e8 |
data/raw/InsTag.tar.zst |
1.051 GiB | 5b28e3ed7f665fa07814ce7135095a791074d41cbbc4b424c116aadd55a43314 |
data/raw/MimicTalk.tar.zst |
0.159 GiB | 995cc475b56b6094a2fe2db78494787dfd6b6a0a281d027a1de5964d6c5e0018 |
data/raw/RADNeRF.tar.zst |
0.451 GiB | 383b1b8b22e1d4be3153a2cfff1c238da6f6c1d6a8799317423d59b9f5598f5a |
data/raw/Real.tar.zst |
3.307 GiB | 691d82bd32224aa5b24cc916e3e28a75974003540c7c3dd8c3d3a7927ae62ea3 |
data/raw/Real3D.tar.zst |
0.167 GiB | 9d65b79d519273dbb57daa25b35dc5805eb47d1a2108fd004515cee3fad723d1 |
data/raw/SyncTalk.tar.zst |
1.369 GiB | 866349ec048bc2fb96008907218fa359e5b6f787fa597ef502fec25bb4e97453 |
data/raw/TalkingGaussian.tar.zst |
0.394 GiB | b120d744b31a054ad8872ed221886e3ee5c9f4bdd8648d8d4673a9599dff4c22 |
Each archive stores paths below 3DDF_Dataset/. Extract the selected archives
into the packaged DeepfakeBench directory:
find data -type f -name '*.tar.zst' -print0 | while IFS= read -r -d '' archive; do
tar --zstd -xf "$archive" -C code/DeepfakeBench
done
The packaged manifests use ./3DDF_Dataset/... paths relative to
code/DeepfakeBench. The original local absolute paths have been removed.
DynTet, Portrait4D-V1, and GaussianHeadAvatar include sanitized DeepfakeBench
JSON metadata. Portrait4D-V2 and Next3D did not have local v6 split JSON files,
so their archive inventory is provided without inventing train/test splits.
Training
The DeepfakeBench entry point is:
cd code/DeepfakeBench
python training/train.py \
--detector_path training/config/detector/altfreezing.yaml \
--ckpt_key 3DDF
Backbone weights and Python environments are not mirrored in this release.
Use the upstream model sources and the dependency files under
code/DeepfakeBench/.
Access And Responsible Use
This repository is gated. It is intended for non-commercial research on deepfake detection, robustness, and provenance. It must not be used for identity impersonation, harassment, surveillance, or creation of deceptive media. Recipients remain responsible for respecting the terms of every source dataset and generation method.
The DeepfakeBench software is distributed under its included CC BY-NC 4.0 license. That software license does not automatically relicense third-party data or model weights. Data components retain their original terms.
Integrity
Run the following from the repository root after download:
sha256sum -c SHA256SUMS
Release composition and byte counts are recorded in
manifests/release_manifest.json.