The DOI matches its Crossref record; 345 of the 345 references it cites were found in a registry or an index; 73 of the 188 sources its claims cite were checked for support.
Countering Malicious DeepFakes: Survey, Battleground, and Horizon
Open accessAbstract
The creation or manipulation of facial appearance through deep generative approaches, known as DeepFake, have achieved significant progress and promoted a wide range of benign and malicious applications, e.g., visual effect assistance in movie and misinformation generation by faking famous persons. The evil side of this new technique poses another popular study, i.e., DeepFake detection aiming to identify the fake faces from the real ones. With the rapid development of the DeepFake-related studies in the community, both sides (i.e., DeepFake generation and detection) have formed the relationship of battleground, pushing the improvements of each other and inspiring new directions, e.g., the evasion of DeepFake detection. Nevertheless, the overview of such battleground and the new direction is unclear and neglected by recent surveys due to the rapid increase of related publications, limiting the in-depth understanding of the tendency and future works. To fill this gap, in this paper, we provide a comprehensive overview and detailed analysis of the research work on the topic of DeepFake generation, DeepFake detection as well as evasion of DeepFake detection, with more than 318 research papers carefully surveyed. We present the taxonomy of various DeepFake generation methods and the categorization of various DeepFake detection methods, and more importantly, we showcase the battleground between the two parties with detailed interactions between the adversaries (DeepFake generation) and the defenders (DeepFake detection). The battleground allows fresh perspective into the latest landscape of the DeepFake research and can provide valuable analysis towards the research challenges and opportunities as well as research trends and future directions. We also elaborately design interactive diagrams (http://www.xujuefei.com/dfsurvey) to allow researchers to explore their own interests on popular DeepFake generators or detectors.
Overview
CoChat Summary
TL;DR
This survey reviews more than 318 papers on DeepFake generation, detection, and evasion, framing generation and detection as an adversarial “battleground.” It provides taxonomies, technical-evolution analyses, dataset summaries, and future research directions for countering malicious DeepFakes.
Key findings
- The survey collected 318 papers; about 78% appeared in the last two years, and it tabulates 83 DeepFake generation methods and 117 DeepFake detection methods.
- DeepFake generation is organized into four categories: entire face synthesis, attribute manipulation, identity swap, and expression swap; identity swap ranks highest in detector popularity, while expression swap is least attempted.
- Detection methods are grouped into spatial-based, frequency-based, and biological-signal-based approaches; evasion methods are grouped into adversarial attacks, frequency-domain trace removal, and filtering/generative reconstruction.
- In challenges, DFDC had 2,114 teams; 60% of submissions had log loss ≤ 0.69, and all top-5 solutions were image-based, with three using EfficientNet. DeeperForensics Challenge 2020 had 115 registered participants and 25 valid submissions; two of the top-3 used EfficientNet and all used augmentation.
Methods
- This is a survey/analysis paper, not an experimental study. Papers were collected from a GitHub repository, keyword matching on Google Scholar, DBLP, and arXiv, plus manual browsing of recent top-tier conferences and workshops.
- The analysis includes taxonomies of generation and detection methods, technical-evolution diagrams, Sankey and chord diagrams for method interactions, Elo ratings for generation methods, and detailed tables of real/fake datasets and detection performance.
Limitations
- The paper states it focuses solely on facial DeepFakes; social and ethical aspects are briefly touched but are not the focus of the survey.
- The paper notes an inherent shortcoming of Elo rating: a generation method that appears once but beats high-score competitors can rise sharply, while a method not detected for a long time keeps an unchanged score that may not reflect its true difficulty.
- The text makes evident that DeepFake detection generally lags behind generation, and many detectors lack robustness, generalization, or explainability; the provided text is also truncated, omitting part of the detection and battleground discussion.
Model-written summary of a shortened version of the paper’s full text (it is longer than we read). Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
62 Valid · 31 Not valid · 108 Not checked · 138 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Although effective, IcGAN can be time-consuming when it aims to perform the manipulation of multiple facial attributes since each attribute is addressed via an independent deep model.”
What the sources say
- Disputedcites Perarnau et al. 2016 sourceExperimental — accuracy not yet measured
The paper describes two independent encoders—one for the latent representation z and one for the attribute information y—not a separate deep model for each facial attribute (Section 4.1, p. 3; Section 6, p. 8). It demonstrates edits across 18 CelebA attributes by modifying y (Section 5.4, p. 7). The copy does not assess whether editing multiple attributes is time-consuming, but it contradicts the claimed reason. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The CramerGANperformssignificantly better than the WGAN.”
What the sources say
- Partly supportedcites Bellemare et al. 2017 sourceExperimental — accuracy not yet measured
Section 5.1 (p. 7) reports that Cramér GAN produced more stable learning and more diverse samples than WGAN-GP; the experiments on p. 8 give examples of both advantages. The claim overgeneralizes this comparison to WGAN and says “significantly better” without specifying the outcomes or conditions the paper evaluated. (LLM, conf=0.86)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“according to the surveys (Verdoliva 2020; Nguyen et al. 2019d; Tolosana et al. 2020; Lyu 2020), GANs are the mainstream baseline technique.”
What the sources say
- Partly supportedcites Verdoliva 2020 sourceExperimental — accuracy not yet measured
Section II, “Fake Content Generation” (p. 3), describes GANs as a successful approach to synthesizing complete media assets, especially photorealistic faces. It also names autoencoders; the copy does not establish that GANs are the mainstream or baseline technique for entire-face synthesis. (LLM, conf=0.79)
- Not matchedcites Nguyen et al. 2019d
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites Tolosana et al. 2020 source
The survey’s “Types of Facial Manipulations” section says entire face synthesis is usually performed through GANs. Its “Manipulation Techniques and Public Databases” section also says the four relevant public databases all use ProGAN or StyleGAN, supporting the characterization of GANs as the mainstream technique for this task. (LLM, conf=0.91)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Karras et al. (2017) propose to grow both the generator and discriminator progressively and implement the PGGAN, which significantly improves the generation size to at least 1024 × 1024.”
What the sources say
- Supportedcites Karras et al. 2017 source
Section 2 (p. 2) describes progressively growing both the generator and discriminator and shows generated images at 1024 × 1024. Section 6.3 (p. 8) says the approach handles high output resolutions robustly and reports 1024 × 1024 images produced by the network. (LLM, conf=0.91)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“In previous works, the absence of paired examples is always the limitation in image transformation tasks. CycleGAN has artfully solved this problem.”
What the sources say
- Supportedcites Zhu et al. 2017 source
The Results section identifies CycleGAN as the authors’ method and says it is applied where paired data does not exist. Section 5.1 reports that it learns image-to-image mappings without paired supervision, supporting the attributed claim. (LLM, conf=0.90)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
90 more claims · 9 with their cited sources checked
Create a free account to see all 339 claims and to verify the 3 references still open.
- Probably not a substantial claim. 191
- It may state two separate claims as one. 101
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Existing surveys [4], [16] relevant to deepfakes predominantly emphasize deepfake generation or general synthetic media detection, whereas our survey is exclusively tailored to the detection of facial deepfakes.”
From Evolving from Single-modal to Multi-modal Facial Deepfake Detection: Progress and Challenges (2024) · cites this paper as [16]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Earlier surveys [16], [17], [18], [19] predominantly address research published up to 2023, missing significant advancements in recent years.”
From Evolving from Single-modal to Multi-modal Facial Deepfake Detection: Progress and Challenges (2024) · cites this paper as [16]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“While previous surveys [4], [16], [17] limit their discussions primarily to visual manipulations, we systematically explore recent developments in detecting multi-modal forgeries that simultaneously involve audio, visual, and textual modalities.”
From Evolving from Single-modal to Multi-modal Facial Deepfake Detection: Progress and Challenges (2024) · cites this paper as [16]
The cited source says what the claim attributes to it.
In the citing paper
“FF++ [57] is one of the most popular datasets [22], but it contains limited types of deepfake data.”
From DEEPFAKER: A Unified Evaluation Platform for Facial Deepfake and Detection Models (2024) · cites this paper as [22]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.7 Fake Dataset
“The DFDC is a publicly-available face swap video dataset, with 128,514 videos, over 100,000 total clips sourced from 3426 actors.”
§ 3.9 Summary of DeepFake Generation Methods
“| DFDC (Dolhansky et al. 2020) | 128,154 | - | - | 23,954 | 104,500 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.6 Real Dataset
“MegaFace(Kemelmacher-Shlizermanetal.2016)includes more than 690K different individuals with one million photos.”
§ 3.9 Summary of DeepFake Generation Methods
“| MegaFace (Kemelmacher-Shlizerman et al. 2016) | 690,572 | 4,700,000 | 6.81 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 2.3 Paper Collection Results
“One fourth of the papers are published in top conferences”
§ 2.3 Paper Collection Results
“Half of the papers are published in top conferences.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 2.1 Survey Scope
“CVPR, which accounts for about half the published top papers.”
§ 2.3 Paper Collection Results
“CVPR, which accounts for about two-thirds the published top papers.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“The DeepFake-TIMIT dataset (Korshunov and Marcel 2018) consists of 620 DeepFake videos of 32 subjects. In DeepFake-TIMIT, each subject has 20 DeepFake videos.”
Reported 620 · recomputed 640
3 more findings
Create a free account to see every finding of this check.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 5×
“FDFtNet (Jeon et al., 2020b) provides a reusable fine-tuning network to improve the capabilities of existing CNN models ( e.g ., SqueezeNet, ShallowNetV3, ResNetV2, and Xception) in detecting fake images effectively.”
MethodInfluentialImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 7×
“On one hand, there will be convergence of many dataset sources into a few very large-scale standardized evaluation datasets for the DeepFake community, similar to the scales of the ImageNet dataset or the COCO object detection dataset (Lin et al., 2014).”
BackgroundMethodInfluentialAn Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al. (2020) · arXiv · doi:10.48550/arxiv.2010.11929 · Cited by 47,472
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460 · cited 1×
“On one hand, there will be convergence of many dataset sources into a few very large-scale standardized evaluation datasets for the DeepFake community, similar to the scales of the ImageNet dataset or the COCO object detection dataset (Lin et al., 2014).”
MethodUnpaired image-to-image translation using cycle-consistent adversarial networks
Zhu, J.-Y., Park, T., Isola, P., & Efros, A. A. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.244 · Cited by 23,602
Xception: Deep Learning with Depthwise Separable Convolutions
Chollet, F. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.195 · Cited by 20,380 · cited 5×
“W a n g ' 21 2017) with significant accuracy reduction on the public datasets (Wang et al., 2020e; Frank et al., 2020).”
BackgroundMethodInfluentialAuto-Encoding Variational Bayes
Kingma, D. P., & Welling, M. (2013) · arXiv · doi:10.48550/arxiv.1312.6114 · Cited by 20,067
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
Tan, M., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1905.11946 · Cited by 17,072 · cited 3×
“Three of the five methods used EfficientNet (Tan and Le, 2019) as the backbone model.”
MethodLearning Transferable Visual Models From Natural Language Supervision
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. (2021) · arXiv · doi:10.48550/arxiv.2103.00020 · Cited by 15,293 · cited 2×
“Examples of vision and language foundational models include Florence (Yuan et al., 2021), CLIP (Radford et al., 2021), ALIGN (Jia et al., 2021), Wu Dao (Beijing Academy of Artificial Intelligence, 2021), etc .”
MethodAdvances in neural information processing systems 7
(1996) · Computers & Mathematics with Applications · doi:10.1016/s0898-1221(96)90463-0 · Cited by 14,281
Research works printed but not linked: 7
- Chen, Z., & Yang, H. (2021). Attentive semantic exploring for manipulated face detection. ICASSP 2021–2021 IEEE international conference on acoustics (pp. 1985–1989). IEEE: Speech and Signal Processing (ICASSP). · DOI
- Friesen, E., & Ekman, P. (1978). Facial action coding system: A technique for the measurement of facial movement. Palo Alto, 3(2), 5.
- Huang, Y., Juefei-Xu, F., Guo, Q., Ma, L., Xie, X., Miao, W., Liu, Y., & Pu, G. (2021a). Dodging DeepFake detection via implicit spatial-domain notch filtering. arXiv preprint arXiv:2009.09213 · arXiv
- Koopman, M., Rodriguez, A. M., & Geradts, Z. (2018). Detection of deepfake video manipulation. In Conference: IMVIP.
- Nhu, T., Na, I., & Kim, S. (2018). Forensics face detection from gans using convolutional neural network. In Proceeding of 2018 international symposium on information technology convergence (ISITC 2018).
- Van Oord, A., Kalchbrenner, N., & Kavukcuoglu, K. (2016). Pixel recurrent neural networks. In: International Conference on Machine Learning, PMLR, pp 1747–1756
- Zhai, L., Juefei-Xu, F., Guo, Q., Xie, X., Ma, L., Feng, W., Qin, S., & Liu, Y. (2022). Adversarial rain attack and defensive deraining for dnn perception. arXiv preprint arXiv:2009.09205 · arXiv
Cited by
Deepfake video detection: challenges and opportunities
Kaur, A., Noori Hoshyar, A., Saikrishna, V., Firmin, S., & Xia, F. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-10810-6 · Cited by 214
UVCGAN: UNet Vision Transformer cycle-consistent GAN for unpaired image-to-image translation
Torbunov, D., Huang, Y., Yu, H., Huang, J., Yoo, S., Lin, M., Viren, B., & Ren, Y. (2023) · 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv56688.2023.00077 · Cited by 189
“Thankfully, there are countermeasures and detection tools [39] developed to counter such misuse.”
BackgroundSelf-citationMasked Relation Learning for DeepFake Detection
Yang, Z., Liang, J., Xu, Y., Zhang, X., & He, R. (2023) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2023.3249566 · Cited by 185
“terfeit faces of images or videos [3], [4], [5], [6].”
BackgroundDeepfakes: current and future trends
Gambín, Á. F., Yazidi, A., Vasilakos, A., Haugerud, H., & Djenouri, Y. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-023-10679-x · Cited by 145
Das, S., Seferbekov, S., Datta, A., Islam, M. S., & Amin, M. R. (2021) · 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) · doi:10.1109/iccvw54120.2021.00421 · Cited by 107
“A survey of multiple DeepFake detection architectures and their performance comparison has been presented in [20].”
Metadata
- Authors
- Juefei-Xu, F., Wang, R., Huang, Y., Guo, Q., Ma, L., & Liu, Y.
- Affiliations
- Alibaba Group (United States) (US)
- Wuhan University (CN)
- East China Normal University (CN)
- Tianjin University (CN)
- University of Alberta (CA)
- Zhejiang Sci-Tech University (CN)
- Corresponding author
- Wang, Run
- Guo, Qing
- Venue
- International Journal of Computer Vision
- Published
- 2022-05-04
- Volume
- 130
- Issue
- 7
- Pages
- 1678-1734
- ISSN
- 0920-5691 (ISSN-L)
- 1573-1405
- PMC
- PMC9066404
- OpenAlex
- W3135485589
- Cited by
- 191
- References
- 325
- Citations by year
- 2026: 23 · 2025: 53 · 2024: 53 · 2023: 43 · 2022: 11
- FWCI
- 13.09
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Licence
- www.springernature.com/gp/researchers/text-and-data-mining (tdm)
- www.springernature.com/gp/researchers/text-and-data-mining (vor)
- Locations
- International Journal of Computer Vision (publishedVersion)
- PubMed Central (publishedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- DR-NTU (Nanyang Technological University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Research Foundation Singapore (NRF2018NCR-NCR005-0001, NRF2018NCR-NSOE003-0001)
- 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
- deepfakes, deepfake detection, deepfake generation, adversarial interaction, misinformation, face manipulation, generative adversarial networks
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (prompt extract-v3).
A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Countering Malicious DeepFakes: Survey, Battleground, and Horizon. https://app.cochat.ai/papers/10.1007/s11263-022-01606-8
- The paper
- Juefei-Xu, F., Wang, R., Huang, Y., Guo, Q., Ma, L., & Liu, Y. (2022). Countering Malicious DeepFakes: Survey, Battleground, and Horizon. International Journal of Computer Vision. https://doi.org/10.1007/s11263-022-01606-8
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.