KoDF: A Large-scale Korean DeepFake Detection Dataset
Open accessAbstract
A variety of effective face-swap and face-reenactment methods have been publicized in recent years, democratizing the face synthesis technology to a great extent. Videos generated as such have come to be called deepfakes with a negative connotation, for various social problems they have caused. Facing the emerging threat of deepfakes, we have built the Korean DeepFake Detection Dataset (KoDF), a large-scale collection of synthesized and real videos focused on Korean subjects. In this paper, we provide a detailed description of methods used to construct the dataset, experimentally show the discrepancy between the distributions of KoDF and existing deepfake detection datasets, and underline the importance of using multiple datasets for real-world generalization. KoDF is publicly available at https://moneybrain-research.github.io/kodf in its entirety (i.e. real clips, synthesized clips, clips with adversarial attack, and metadata).
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Most available datasets provide examples of face manipulations through either face swapping or face reenactment techniques.
“Most of the available datasets provide examples of face manipulations through either face swapping [16, 34, 77] or face reenactment [31, 35] techniques.”
From AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset (2024) · cites this paper as [35]
The Korean DeepFake Detection Dataset contains 37,942 videos of paid subjects, including 62,166 real videos and 175,776 fake ones created using six methods.
“Korean DeepFake Detection Dataset (Kwon et al., 2021): This dataset contains 37,942 videos of paid subjects (395 Koreans and 8 Southeastern Asians), including 62,166 real videos and 175,776 fake ones created using six methods”
From Deepfake: definitions, performance metrics and standards, datasets, and a meta-review (2024) · cites this paper as Kwon et al., 2021
Several large-scale datasets such as DFDC, DeeperForensics, KoDF, and DF-Platter have been proposed.
“With the rapid advancements in content generation technology, several large-scale datasets such as DFDC [16], DeeperForensics [28], KoDF [35], and DF-Platter [41] have been proposed.”
From AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset (2024) · cites this paper as [35]
KoDF comprises 62,166 authentic videos and 175,776 fake videos, featuring 403 different subjects.
“This dataset comprises 62,166 authentic videos and 175,776 fake videos, featuring 403 different subjects.”
From Deepfake Attacks: Generation, Detection, Datasets, Challenges, and Research Directions (2023) · cites this paper as [129]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 39 · Cited by 145
Research works printed but not linked: 12
- Shruti Agarwal, Hany Farid, Yuming Gu, Mingming He, Koki Nagano, and Hao Li. Protecting world leaders against deep fakes. IEEE Conference on Computer Vision and Pat- tern Recognition (CVPR) Workshops, pages 38-45, 2019.
- Gereon Fox, Wentao Liu, Hyeongwoo Kim, Hans-Peter Sei- del, Mohamed Elgharib, and Christian Theobalt. Vide- oforensicshq: Detecting high-quality manipulated face videos. arXiv preprint arXiv:2005.10360, 2020. · arXiv
- Dou Goodman, Hao Xin, Wang Yang, Wu Yuesheng, Xiong Junfeng, and Zhang Huan. Advbox: A toolbox to gener- ate adversarial examples that fool neural networks. arXiv preprint arXiv:2001.05574, 2020. · arXiv
- Karen Hao. Deepfake porn is ruining women's lives. now the law may finally ban it. MIT Technology Review, 2021.
- Yihao Huang, Felix Juefei-Xu, Qing Guo, Xiaofei Xie, Lei Ma, Weikai Miao, Yang Liu, and Geguang Pu. Fakeretouch: Evading deepfakes detection via the guidance of deliberate noise. arXiv preprint arXiv:2009.09213, 2020. · arXiv
- Iacopo Masi, Aditya Killekar, Royston Mascarenhas, Shenoy Gurudatt, and Wael AbdAlmageed. Two-branch re- current network for isolating deepfakes in videos. Computer Vision -ECCV 2020), pages 667-684, 2020.
- Paarth Neekhara, Brian Dolhansky, Joanna Bitton, and Cris- tian Ferrer. Adversarial threats to deepfake detection: A practical perspective. arXiv preprint arXiv:2011.09957, 2020. · arXiv
- Ekraam Sabir, Jiaxin Cheng, Ayush Jaiswal, Wael AbdAl- mageed, Iacopo Masi, and Prem Natarajan. Recurrent convo- lutional strategies for face manipulation detection in videos. IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR) Workshops, pages 80-87, 2019.
- Aliaksandr Siarohin, Stéphane Lathuilière, Sergey Tulyakov, Elisa Ricci, and Nicu Sebe. First order motion model for im- age animation. Advances in Neural Information Processing Systems (NeurIPS), 2019.
- Catherine Stupp. Fraudsters used ai to mimic ceo's voice in unusual cybercrime case. The Wall Street Journal, 2019.
- Ruben Tolosana, Sergio Romero-Tapiador, Julian Fierrez, and Ruben Vera-Rodriguez. Deepfakes evolution: Analy- sis of facial regions and fake detection performance. Pattern Recognition. ICPR International Workshops and Challenges, pages 442-456, 2021.
- Ran Yi, Zipeng Ye, Juyong Zhang, Hujun Bao, and Yongjin Liu. Audio-driven talking face video generation with learning-based personalized head pose. arXiv preprint arXiv:2002.10137, 2020. · arXiv
- Authors
- Kwon, P., You, J., Nam, G., Park, S., & Chae, G.
- Affiliations
- MoneyBrain Inc.,Seoul,Republic of Korea
- Venue
- 2021 IEEE/CVF International Conference on Computer Vision (ICCV)
- Publisher
- IEEE
- Published
- 2021-10
- Pages
- 10724-10733
- OpenAlex
- W3204680838
- Cited by
- 145
- References
- 39
- Citations by year
- 2026: 22 · 2025: 38 · 2024: 42 · 2023: 25 · 2022: 14 · 2021: 4
- FWCI
- 6.94
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE/CVF International Conference on Computer Vision (ICCV) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Ministry of Science and ICT, South Korea
- Institute for Information and Communications Technology Promotion
- National Institute on Aging
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face recognition and analysis
- Keywords
- deepfake detection, face swapping, face reenactment, synthetic video, adversarial attacks, dataset construction, cross-dataset generalization, Korean speakers
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
