Pith. sign in

REVIEW 4 major objections 5 minor 68 references

FakeI2V-Bench: Benchmarking the Applicability of Image-level Deepfake Detectors for Deepfake Video Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read FakeI2V-Bench claims that fine-tuned image-level detectors with random-forest aggregation surpass the best video-level deepfake detector on a 97,548-video benchmark.

desk verdict Useful new benchmark, but the headline comparison is not yet supported: image detectors get test-set-chosen aggregation and extra training data the video detectors don't. read the letter →

arxiv 2608.03096 v1 pith:NPVX7QPH submitted 2026-08-04 cs.CR cs.AIcs.CV

Pei Li , Sihan Chen , Delong Ran , Tianshuo Cong This is my paper · ORCID
classification cs.CRcs.AIcs.CV
keywords deepfakedetectionvideodeepfakesAI-generatedimage-leveldetectorsframe-to-videoaggregationdetectorbenchmarkingAIGCIV-Bridge
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

FakeI2V-Bench argues that image-level deepfake detectors—models that score individual frames—are not obsolete in the video era, but are being evaluated the wrong way. On a new benchmark of 97,548 videos spanning facial forgeries and general AI-generated content, naive image-level detectors drop sharply at frame level, yet one of them reaches 80.16% AUC after simple frame-to-video aggregation, slightly above the best video-level detector's 79.99% AUC. The paper's IV-Bridge pipeline, which fine-tunes image detectors on video frames and then learns to fuse six frame-score statistics with a random forest, lifts eleven of twelve image-level detectors above that video-level state of the art, with the best variant reaching 93.80% AUC. If these comparisons hold, the practical recipe for video deepfake detection becomes cheaper and more general than current video-level models.

What carries the argument

IV-Bridge, a two-stage adaptation pipeline. Stage one, Video-Frame Fine-Tuning (VFT), fine-tunes each image-level detector on a mixed pool of real and fake video frames from FaceForensics++ and the GenVideo training split, using cross-entropy or binary cross-entropy depending on the detector's output head. Stage two, Multi-Mode Aggregation (MMA), computes six frame-score statistics per video—the sampled single frame, mean, max, min, median, and variance—and feeds the six values into a random forest trained on a separate video set to output the final forgery probability. The key work the random forest does is to learn, per detector and per video, which aggregation mode or combination of modes is most trustworthy, replacing a fixed rule such as 'average over frames'.

What would settle it

Re-run the benchmark with one aggregation mode fixed before seeing test labels and with the random forest trained on an explicitly disjoint, published split; if the best adapted image-level detector no longer exceeds FTCN's 79.99% mean AUC, the claimed superiority is an artifact of the evaluation protocol.

Watch

Extended reading notes

Core claim

The central discovery is that the forgery signal needed for video deepfake detection mostly lives in frames, not in long-range temporal structure, and the bottleneck is the frame-to-video decision rule. The paper shows that ten of twelve image-level detectors lose substantial accuracy when moved from static forged images to video frames (e.g., NPR drops from 92.17% to 55.38% AUC), but that aggregation restores much of the loss: RINE goes from 71.11% frame-level AUC to 80.16% video-level AUC, surpassing the strongest video-level detector evaluated (FTCN at 79.99%). IV-Bridge then closes most of the remaining gap by fine-tuning each image detector on real and fake video frames and feeding a six-statistic summary of its frame scores through a random forest. In the paper's Table 3, eleven of twelve enhanced detectors beat FTCN overall, and the best, RINE-IV, reaches 93.80% AUC and 97.63% AP.

Load-bearing premise

The headline advantage is measured by allowing each image-level detector to be scored with whichever of six aggregation modes happens to look best on the test set, and by training the random-forest aggregator on data whose separation from the evaluation videos is not spelled out.

Editorial extensions

If this is right

  • Eleven of the twelve IV-Bridge-enhanced image-level detectors exceed the best video-level detector (FTCN, 79.99% mean AUC), with RINE-IV reaching 93.80% AUC and 97.63% AP.
  • The practical cost equation flips: three enhanced image-level detectors (NPR-IV, Patch-IV, CoDE-IV) beat FTCN on accuracy, parameter count, and inference time at once, with Patch-IV at 4.34M parameters and 9.69 ms per video.
  • Comparing image-level detectors at frame level against video-level detectors is misleading; once frame scores are aggregated to a video decision, even a naive detector (RINE, 80.16% AUC) can edge past the strongest video-level model.
  • Adapted image-level detectors generalize across 14 video generation models, averaging 95.51% AUC versus 91.01% for the best video-level detector, while Sora and image-initialized generators (SVD, MuseV) remain the hardest cases.
  • Under sparse tampering (only 20% forged frames in Celeb-DF v2), all detectors degrade, but the best enhanced image-level detector (Patch-IV, 77.15% AUC) still clearly beats the best video-level detector (LAA-Net, 67.15% AUC).

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension the authors leave implicit: the six frame-score statistics are detector-agnostic, so one shared random forest could be trained on scores from all twelve detectors at once; the resulting ensemble might exceed any single -IV variant.
  • Because the aggregation features encode only score spread, not score order, an extension worth testing is adding temporal-order statistics—where in the video the max, min, or highest-variance segment occurs—to address the sparse-tampering regime where all current variants lose ground.
  • The reported best-of-six protocol means the numbers should be read as an upper bound on what a deployed system would achieve; fixing one aggregation mode on a validation split would give the operational performance estimate.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces FakeI2V-Bench, a benchmark of 97,548 videos (CDFV2, DFD, GenVideo, GenVidBench) for evaluating deepfake video detectors, with a focus on systematically assessing image-level detectors. It evaluates eight video-level detectors and twelve naive image-level detectors, reporting frame-level and video-level performance under six aggregation modes. The paper further proposes IV-Bridge, a two-stage enhancement consisting of Video-Frame Fine-Tuning (VFT) and Multi-Mode Aggregation (MMA) via a random forest, claiming that eleven of twelve IV-Bridge-enhanced detectors surpass the best video-level detector (FTCN, 79.99% AUC), with RINE-IV reaching 93.80% AUC. The central claims are that naive image-level detectors can slightly surpass video-level detectors after test-set-optimal aggregation (80.16% vs. 79.99% AUC) and that IV-Bridge-enhanced image-level detectors are both more accurate and cheaper than video-level detectors.

Significance. If the headline results were established under a sound protocol, the benchmark would be a valuable resource: it is the first to systematically compare a broad set of image-level detectors on video-level tasks, it covers recent generation models, and it proposes a lightweight enhancement pathway (IV-Bridge) that could make mature image-level detectors practical for video. The release of code and data is a concrete strength, and the per-detector, per-dataset breakdown in Table 3 is useful for the community. However, the two central quantitative claims rest on protocol choices that currently overstate the case: reporting the post-hoc best aggregation mode for naive image-level detectors and failing to specify the training set for the random forest in MMA. These issues are load-bearing for the paper's main comparisons, though they appear fixable with a validation-based protocol and full disclosure.

major comments (4)
  1. [Section 5.2 and Table 3 caption] The paper reports naive image-level detector results as 'the optimal aggregated results by one of the six modes' (Table 3 caption). Because the aggregation mode is selected after observing test-set performance, the 80.16% AUC for RINE is the maximum over six correlated statistics rather than the performance of a fixed, independently selected pipeline. The 0.17 percentage-point margin over FTCN (79.99%) is exactly the kind of number that this selection process can manufacture. Please fix the aggregation protocol on a validation split (or pre-register a single mode per detector and report all six modes for transparency), then report the test-set numbers under that fixed protocol.
  2. [Section 6.1 (MMA) and Section 6.2] The random forest classifier C_RF in MMA is trained on a dataset D_vid that is never specified. The paper explicitly states that the fine-tuning set D_ft has no overlap with the evaluation set, but it is silent about D_vid. If D_vid contains FakeI2V-Bench evaluation videos or frames, the headline 93.80% AUC for RINE-IV is directly contaminated. Please define D_vid, including its source, size, and how it relates to the evaluation sets, and rule out any overlap. Ideally, train the random forest via cross-validation on a disjoint split so the aggregation learner has genuinely unseen data.
  3. [Section 6.2 and Table 5] The comparison between IV-Bridge-enhanced image-level detectors and video-level detectors is asymmetric: image-level detectors receive VFT fine-tuning on FF++ (c23) plus GenVideo training frames and a learned random-forest aggregator, while all video-level detectors are evaluated with official pretrained weights and no adaptation. The deployment-cost table (Table 5) therefore compares 'off-the-shelf video-level detectors' against 'fine-tuned and learned-aggregated image-level detectors,' which is not a like-for-like comparison. Please either apply an equivalent fine-tuning/adaptation protocol to the video-level detectors, or explicitly restrict the claim to 'image-level detectors augmented by IV-Bridge fine-tuning and aggregation.'
  4. [Tables 3 and 5; Appendix A] No error bars, confidence intervals, or training seeds are reported, and the SMP aggregation mode in Appendix A is explicitly random. Given that the naive-image-level versus video-level margin is only 0.17 AUC points and that the SMP mode is stochastic, a single run is insufficient to support the comparison. Please report means and standard deviations over at least three seeds, or otherwise fix and state the random seed and show that the conclusions are stable.
minor comments (5)
  1. [Section 5.2 and Table 3 caption] The phrase 'the optimal aggregated results by default' is ambiguous: it is not clear whether the optimal mode is chosen per detector globally or per dataset, and the caption should state this explicitly.
  2. [Table 5] The 'Time (ms)' column should specify what workload the timing refers to (e.g., one video of a given length, one frame, or one full evaluation clip), since inference time for frame-based methods depends on the number of frames processed.
  3. [Section 6.2] The fine-tuning details for VFT are incomplete: learning rate, number of epochs, optimizer, and batch size are not given, although the paper states it follows original hyperparameters. Please provide these in an appendix or supplementary material.
  4. [Figures 2, 4, 5, 6, and 7] In the provided version, the figure labels appear as corrupted glyph sequences (e.g., '/uni00000013'); if these are not rendering artifacts of the PDF, the figures need to be regenerated with correct text labels.
  5. [Abstract and Table 2] The number '97, 548' appears with an extra space in the abstract; please fix the formatting. Also, consider adding the total real/fake video counts in Table 2 for clarity.

Circularity Check

1 steps flagged · score 6.0 of 10

Naive image-level results are oracle-selected over six test-set aggregation modes, and IV-Bridge's MMA random forest trains on an unspecified D_vid whose separation from the evaluation set is never stated; the headline 80.16% and 93.80% numbers are thus partly self-confirming rather than independent predictions.

  1. fitted input called prediction [Section 5.2, Table 3 caption]
    "Table 3 caption: 'The results of naive image-level detectors are the optimal aggregated results by one of the six modes.' Section 5.2: 'The reported results are the optimal aggregated results by default.' 'For example, RINE achieves an AUC of only 68.60% under the VAR mode, whereas its performance improves to 80.16% when the MIN strategy is adopted.'"

    The reported naive image-level AUC is not the test performance of a fixed detector-plus-aggregation pipeline; it is the maximum over six frame-to-video aggregation modes, all evaluated on the same FakeI2V-Bench test set with no validation split described. Selecting the mode after observing test AUC makes 80.16% an order statistic of test-set measurements, so the abstract's claim that the best image-level detector (80.16%) slightly outperforms FTCN (79.99%) is an oracle comparison, not a matched head-to-head. The paper's own RINE example (VAR 68.60% vs MIN 80.16%) shows the mode choice is data-dependent, so the reported 'prediction' is statistically forced by the selection procedure.

full rationale

FakeI2V-Bench is otherwise a self-contained empirical benchmark: frame-level scores come from external detector weights, datasets are public, and per-mode results are reported. There is no equation-level circularity, no self-citation chain, and no imported uniqueness theorem. The central circularity is the test-set aggregation-mode selection in Table 3: the naive image-level AUCs are reported as the best of six modes evaluated on the same test set, with no validation split described, so the claim that an image-level detector 'slightly outperforms' FTCN is an oracle comparison rather than a fixed-pipeline prediction. A further missing-support flag, not treated as a demonstrated circular step, is that IV-Bridge's MMA stage trains a random forest on an unspecified D_vid; the paper's non-overlap statement covers only the VFT data D_ft, so it never rules out that the aggregation classifier behind the 93.80% RINE-IV result was trained on evaluation videos or frames. If that overlap existed, the headline IV-Bridge result would be a fitted value on the test set. These are selection and potential-leakage risks, not algebraic self-reference, and the paper is transparent that the naive results are 'optimal aggregated results by default.' A score of 6 reflects that the headline superiority claims are partly manufactured by test-set mode selection and by an unverified training-set boundary for the aggregation classifier, while the underlying benchmark measurements remain external and potentially valuable.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

No new physical or ontological entities are introduced. The empirical claims rest on dataset curation, pretrained weights, and evaluation protocol choices; the main ledger entries are the post-hoc aggregation mode selection and unspecified MMA training data.

free parameters (3)
  • Optimal aggregation mode selection for each naive image-level detector = e.g., MIN for RINE; best of SMP/AVG/MAX/MIN/MED/VAR per detector
    Table 3 caption reports naive results as 'the optimal aggregated results by one of the six modes'; choosing the mode after seeing test labels is a free choice fitted to the evaluation data.
  • Random forest hyperparameters in MMA = up to 800 trees, max depth 8
    Hand-selected configuration; no ablation or cross-validation is reported to show the result is insensitive to these choices.
  • VFT fine-tuning data composition = 197,533 real and 207,476 fake frames from FF++ and GenVideo training split
    Authors chose these sources; they may favor generators similar to the evaluation data and are not controlled against video-level baselines that receive no fine-tuning.
assumptions (4)
  • domain assumption Official pretrained weights for evaluated detectors correctly represent each published method.
    Sections 3.2 and 3.3 rely on official weights except DeMamba and LNP, which are retrained; weight provenance and retraining fidelity are not independently verified.
  • domain assumption The fine-tuning and MMA training data do not overlap FakeI2V-Bench evaluation videos.
    Stated only for VFT frames in Section 6.2; D_vid for MMA is not described, so this assumption is unverified for the aggregation stage.
  • domain assumption Labels in CDFV2, DFD, GV test split, and GVB subset are correct and representative of real versus AI-generated video.
    The benchmark inherits labels from prior datasets; no manual re-verification or label audit is presented.
  • domain assumption Mean AUC across four heterogeneous datasets is an appropriate single-number comparison metric.
    Headline comparisons use the average of per-dataset AUCs without confidence intervals; aggregation may hide large per-dataset variation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FakeI2V-Bench: Benchmarking the Applicability of Image-level Deepfake Detectors for Deepfake Video Detection." pith.science (2026). https://pith.science/paper/NPVX7QPH

@misc{pith2026260803096,
  author       = {Pith},
  title        = {Pith review of: FakeI2V-Bench: Benchmarking the Applicability of Image-level Deepfake Detectors for Deepfake Video Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NPVX7QPH}},
  note         = {Machine review of arXiv:2608.03096}
}
read the original abstract

Recent advances in video generation models have significantly intensified the deepfake threat, yet the current deepfake video detection benchmarks remain underdeveloped. In particular, the effectiveness of image-level detectors in the video domain has not been systematically assessed. To fill this gap, we present FakeI2V-Bench, a benchmark for evaluating state-of-the-art video-level deepfake detectors in challenging scenarios, with a particular focus on systematically assessing the performance of image-level deepfake detectors in the video domain. FakeI2V-Bench comprises 97,548 videos, containing content generated by the latest powerful generation models and covering a broader range of categories. Using this dataset, we conduct a systematic evaluation of eight video-level detectors and twelve representative image-level detectors. Experimental results show that the best-performing image-level detector achieves an 80.16% AUC, slightly outperforming the strongest video-level detector (i.e., 79.99% AUC). Going beyond benchmarking, we present IV-Bridge, a general framework that enhances the applicability of image-level deepfake detectors to videos. IV-Bridge employs a random forest model with statistical features to aggregate frame-level predictions, allowing eleven image-level detectors to surpass state-of-the-art video-level approaches, with the best-performing variant achieving a 93.80% AUC. Overall, FakeI2V-Bench establishes a rigorous benchmark for deepfake video detection and introduces a novel pathway for extending image-level detectors to the video domain, offering new insights and directions for future research. Code and data are available at https://github.com/CryptoAILab/FakeI2V-Bench.

Figures

Figures reproduced from arXiv: 2608.03096 by the authors.

Figure 1
Figure 1. An overview of FakeI2V-Bench. FakeI2V-Bench is designed to comprehensively evaluate the performance of deepfake [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The video-level performances of naive image-level detectors using six aggregation modes. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Frame-level detection performance (AUC %) of image-level detectors on FakeI2V-Bench. “w/o VFT” denotes the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Results of cross-model detection performance (AUC [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Results of cross-model detection AUC (%) on GV [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Results of cross-model detection AUC (%) on GVB [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: The video-level performances of eight naive image-level detectors using six aggregation modes. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 35 canonical work pages

  1. [1]

    Lorenzo Baraldi, Federico Cocchi, Marcella Cornia, Lorenzo Baraldi, Alessandro Nicolosi, and Rita Cucchiara. 2024. Contrasting Deepfakes Diffusion via Con- trastive Learning and Global-Local Similarities. InSpringer European Conference on Computer Vision (ECCV)

    work page 2024

  2. [2]

    Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram Voleti, Adam Letts, Varun Jampani, and Robin Rombach. 2023. Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets. arXiv Preprint 2311.15127. arXiv:2311.15127 https://arxiv.org/abs/2311.15127

    arXiv 2023

  3. [3]

    Matyas Bohacek and Hany Farid. 2024. The making of an AI news anchor—and its implications.Proceedings of the National Academy of Sciences121, 1 (2024), e2315678121

    work page 2024

  4. [4]

    Leo Breiman. 2001. Random Forests.Mach. Learn.45, 1 (2001), 5–32. doi:10.1023/A: 1010933404324

    doi:10.1023/a: 2001

  5. [5]

    Andrew Brock, Jeff Donahue, and Karen Simonyan. 2018. Large scale GAN training for high fidelity natural image synthesis.arXiv preprint arXiv:1809.11096 (2018)

    arXiv 2018

  6. [6]

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luhman, Eric Luhman, Clarence Ng, Ricky Wang, and Aditya Ramesh. 2024. Video generation models as world simulators. https: //openai.com/index/sora/. Accessed: 2025-04-08

    work page 2024

  7. [7]

    Lucy Chai, David Bau, Ser-Nam Lim, and Phillip Isola. 2020. What makes fake im- ages detectable? Understanding properties that generalize. InSpringer European Conference on Computer Vision (ECCV)

    work page 2020

  8. [8]

    Baoying Chen, Jishen Zeng, Jianquan Yang, and Rui Yang. 2024. DRCT: Diffusion Reconstruction Contrastive Training towards Universal Detection of Diffusion Generated Images. InForty-first International Conference on Machine Learning. https://openreview.net/forum?id=oRLwyayrh1

    work page 2024

Show all 68 references
  1. [9]

    Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. 2018. Learning to see in the dark. InProceedings of the IEEE conference on computer vision and pattern recognition. 3291–3300

    2018

  2. [10]

    Haoxing Chen, Yan Hong, Zizheng Huang, Zhuoer Xu, Zhangxuan Gu, Yaohui Li, Jun Lan, Huijia Zhu, Jianfu Zhang, Weiqiang Wang, and Huaxiong Li. 2024. DeMamba: AI-Generated Video Detection on Million-Scale GenVideo Benchmark. arXiv Preprint 2405.19707. arXiv:2405.19707 https://ar...

    2024 arXiv

  3. [11]

    Qifeng Chen and Vladlen Koltun. 2017. Photographic image synthesis with cascaded refinement networks. InProceedings of the IEEE international conference on computer vision. 1511–1520. KDD 2026, August 9–13, 2026, Jeju Island, Republic of Korea. Pei Li, Sihan Chen, Delong Ran, ...

    2017

  4. [12]

    Yunjey Choi, Minje Choi, Munyoung Kim, Jung-Woo Ha, Sunghun Kim, and Jaegul Choo. 2018. Stargan: Unified generative adversarial networks for multi- domain image-to-image translation. InProceedings of the IEEE conference on computer vision and pattern recognition. 8789–8797

    2018

  5. [13]

    Riccardo Corvi, Davide Cozzolino, Giada Zingarini, Giovanni Poggi, Koki Nagano, and Luisa Verdoliva. 2023. On The Detection of Synthetic Images Generated by Diffusion Models. InIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

    2023

  6. [14]

    Tao Dai, Jianrui Cai, Yongbing Zhang, Shu-Tao Xia, and Lei Zhang. 2019. Second- order attention network for single image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 11065–11074

    2019

  7. [15]

    Jingyi Deng, Chenhao Lin, Pengbin Hu, Chao Shen, Qian Wang, Qi Li, and Qiming Li. 2024. Towards benchmarking and evaluating deepfake detection. IEEE Transactions on Dependable and Secure Computing21, 6 (2024), 5112–5127

    2024

  8. [16]

    Prafulla Dhariwal and Alexander Nichol. 2021. Diffusion models beat gans on image synthesis.Advances in neural information processing systems34 (2021), 8780–8794

    2021

  9. [17]

    Google and Jigsaw. 2019. Contributing Data to Deepfake Detection Research. https://research.google/blog/contributing-data-to-deepfake-detection- research/. Accessed: 2025-04-10

    2019

  10. [18]

    Devin Gordon. 2024. What if A.I. Is Actually Good for Hollywood? https: //www.nytimes.com/2024/11/01/magazine/ai-hollywood-movies-cgi.html. Accessed: 2024-12-16

    2024

  11. [19]

    Wenyi Hong, Ming Ding, Wendi Zheng, Xinghan Liu, and Jie Tang. 2022. CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transform- ers. arXiv Preprint 2205.15868. arXiv:2205.15868 https://arxiv.org/abs/2205.15868

    2022 arXiv

  12. [20]

    Christian Internò, Robert Geirhos, Markus Olhofer, Sunny Liu, Barbara Hammer, and David Klindt. 2026. AI-Generated Video Detection via Perceptual Straight- ening. arXiv:2507.00583 [cs.CV] https://arxiv.org/abs/2507.00583

    2026

  13. [21]

    Chaewon Kang, Seoyoon Jeong, Jonghyun Lee, Daejin Choi, Simon S Woo, and Jinyoung Han. 2025. HiDF: A Human-Indistinguishable Deepfake Dataset. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. 5527–5538

    2025

  14. [22]

    Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. 2017. Progressive growing of gans for improved quality, stability, and variation.arXiv preprint arXiv:1710.10196(2017)

    2017 arXiv

  15. [23]

    Tero Karras, Samuli Laine, and Timo Aila. 2019. A style-based generator ar- chitecture for generative adversarial networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 4401–4410

    2019

  16. [24]

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. 2020. Analyzing and improving the image quality of stylegan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 8110–8119

    2020

  17. [25]

    Dan Kondratyuk, Lijun Yu, Xiuye Gu, José Lezama, Jonathan Huang, Grant Schindler, Rachel Hornung, Vighnesh Birodkar, Jimmy Yan, Ming-Chang Chiu, et al. 2023. Videopoet: A large language model for zero-shot video generation. arXiv preprint 2312.14125. (2023)

    2023 arXiv

  18. [26]

    Christos Koutlis and Symeon Papadopoulos. 2024. Leveraging Representations from Intermediate Encoder-Blocks for Synthetic Image Detection. InSpringer European Conference on Computer Vision (ECCV)

    2024

  19. [27]

    Ke Li, Tianhao Zhang, and Jitendra Malik. 2019. Diverse image synthesis from semantic layouts via conditional imle. InProceedings of the IEEE/CVF International Conference on Computer Vision. 4220–4229

    2019

  20. [28]

    Yuezun Li, Xin Yang, Pu Sun, Honggang Qi, and Siwei Lyu. 2020. Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2020

  21. [29]

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Microsoft coco: Common objects in context. InEuropean conference on computer vision. Springer, 740–755

    2014

  22. [30]

    Bo Liu, Fan Yang, Xiuli Bi, Bin Xiao, Weisheng Li, and Xinbo Gao. 2022. Detecting Generated Images by Real Images. InSpringer European Conference on Computer Vision (ECCV)

    2022

  23. [31]

    Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. 2015. Deep learning face attributes in the wild. InProceedings of the IEEE international conference on computer vision. 3730–3738

    2015

  24. [32]

    Dat Nguyen, Nesryne Mejri, Inder Pal Singh, Polina Kuleshova, Marcella Astrid, Anis Kacem, Enjie Ghorbel, and Djamila Aouada. 2024. LAA-Net: Localized Artifact Attention Network for Quality-Agnostic and Generalizable Deepfake Detection. InIEEE/CVF Conference on Computer Vision...

    2024

  25. [33]

    Zhenliang Ni, Qiangyu Yan, Mouxiao Huang, Tianning Yuan, Yehui Tang, Hailin Hu, Xinghao Chen, and Yunhe Wang. 2025. GenVidBench: A Challenging Benchmark for Detecting AI-Generated Video. arXiv Preprint 2501.11340. arXiv:2501.11340 https://arxiv.org/abs/2501.11340

    2025

  26. [34]

    Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. 2021. Glide: Towards photorealistic image generation and editing with text-guided diffusion models.arXiv preprint arXiv:2112.10741(2021)

    2021 arXiv

  27. [35]

    Superior Court of the State of Washington for King County. 2024. FINDINGS OF FACT AND CONCLUSIONS OF LAW RE: FRYE HEARING ON ADMISSIBILITY OF VIDEOS ENHANCED BY ARTIFICIAL INTELLIGENCE. https://fingfx.thom sonreuters.com/gfx/legaldocs/zgvokxekavd/04192024ai_wash.pdf. Accessed:...

    2024

  28. [36]

    Utkarsh Ojha, Yuheng Li, and Yong Jae Lee. 2023. Towards Universal Fake Image Detectors that Generalize Across Generative Models. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2023

  29. [37]

    Yan Pang, Yang Zhang, and Tianhao Wang. 2024. VGMShield: Mitigating Misuse of Video Generative Models. arXiv Preprint 2402.13126. arXiv:2402.13126 https: //arxiv.org/abs/2402.13126

    2024 arXiv

  30. [38]

    Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun-Yan Zhu. 2019. Semantic image synthesis with spatially-adaptive normalization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2337–2346

    2019

  31. [39]

    Yuyang Qian, Guojun Yin, Lu Sheng, Zixuan Chen, and Jing Shao. 2020. Think- ing in frequency: Face forgery detection by mining frequency-aware clues. In European conference on computer vision. Springer, 86–103

    2020

  32. [40]

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever. 2021. Zero-shot text-to-image generation. InInternational conference on machine learning. Pmlr, 8821–8831

    2021

  33. [41]

    Runway Research. 2023. Gen-2: Generate novel videos with text, images or video clips. https://runwayml.com/research/gen-2. Accessed: 2025-04-08

    2023

  34. [42]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-Resolution Image Synthesis With Latent Diffusion Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 10684–10695

    2022

  35. [43]

    Kevin Roose. 2018. Here Come the Fake Videos, Too. https://www.nytimes.com/ 2018/03/04/technology/fake-videos-deepfakes.html. Accessed: 2025-08-27

    2018

  36. [44]

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al

  37. [45]

    Andreas Rössler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, and Matthias Niessner. 2019. FaceForensics++: Learning to Detect Manipulated Facial Images. In2019 IEEE/CVF International Conference on Computer Vision (ICCV). 1–11. doi:10.1109/ICCV.2019.00009

    2019

  38. [46]

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki

  39. [47]

    Zeyang Sha, Zheng Li, Ning Yu, and Yang Zhang. 2023. DE-FAKE: Detection and Attribution of Fake Images Generated by Text-to-Image Generation Model. In ACM SIGSAC Conference on Computer and Communications Security (CCS)

    2023

  40. [48]

    Xiufeng Song, Xiao Guo, Jiache Zhang, Qirui Li, Lei Bai, Xiaoming Liu, Guangtao Zhai, and Xiaohong Liu. 2024. On Learning Multi-Modal Forgery Representation for Diffusion Generated Video Detection. InConference on Neural Information Processing Systems (NeurIPS)

    2024

  41. [49]

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, Ping Liu, and Yunchao Wei. 2024. Rethinking the Up-Sampling Operations in CNN-based Generative Net- work for Generalizable Deepfake Detection. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2024

  42. [50]

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, and Yunchao Wei

  43. [51]

    Justus Thies, Michael Zollhöfer, Marc Stamminger, Christian Theobalt, and Matthias Nießner. 2016. Face2Face: Real-time Face Capture and Reenactment of RGB Videos. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2016

  44. [52]

    Jiuniu Wang, Hangjie Yuan, Dayou Chen, Yingya Zhang, Xiang Wang, and Shiwei Zhang. 2023. Modelscope text-to-video technical report. arXiv Preprint 2308.06571. arXiv:2308.06571 https://arxiv.org/abs/2308.06571

    2023 arXiv

  45. [53]

    Sheng-Yu Wang, Oliver Wang, Richard Zhang, Andrew Owens, and Alexei A Efros. 2020. CNN-generated images are surprisingly easy to spot... for now. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2020

  46. [54]

    Zhendong Wang, Jianmin Bao, Wengang Zhou, Weilun Wang, Hezhen Hu, Hong Chen, and Houqiang Li. 2023. DIRE for Diffusion-Generated Image Detection. In IEEE/CVF International Conference on Computer Vision (ICCV)

    2023

  47. [55]

    Zhendong Wang, Jianmin Bao, Wengang Zhou, Weilun Wang, and Houqiang Li

  48. [56]

    Zhiqiang Xia, Zhaokang Chen, Bin Wu, Chao Li, Kwok-Wai Hung, Chao Zhan, Yingjie He, and Wenjiang Zhou. 2024. MuseV: Infinite-length and High Fidelity Virtual Human Video Generation with Visual Conditioned Parallel Denoising. https://tmelyralab.github.io/MuseV_Page/. Accessed: ...

    2024

  49. [57]

    Jun Xu, Tao Mei, Ting Yao, and Yong Rui. 2016. Msr-vtt: A large video description dataset for bridging video and language. InProceedings of the IEEE conference on computer vision and pattern recognition. 5288–5296. FakeI2V-Bench: Benchmarking the Applicability of Image-level D...

    2016

  50. [58]

    Zhiyuan Yan, Yong Zhang, Xinhang Yuan, Siwei Lyu, and Baoyuan Wu. 2023. DeepfakeBench: a comprehensive benchmark of deepfake detection. InProceed- ings of the 37th International Conference on Neural Information Processing Systems (New Orleans, LA, USA)(NIPS ’23). Curran Associ...

    2023

  51. [59]

    InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    AltFreezing for More General Video Face Forgery Detection. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

  52. [60]

    Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. 2015. Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.arXiv preprint arXiv:1506.03365(2015)

    2015 arXiv

  53. [61]

    Chende Zheng, Ruiqi Suo, Chenhao Lin, Zhengyu Zhao, Le Yang, Shuai Liu, Minghui Yang, Cong Wang, and Chao Shen. 2025. D3: Training-free ai-generated video detection using second-order features. InProceedings of the IEEE/CVF International Conference on Computer Vision. 12852–12862

    2025

  54. [62]

    Yinglin Zheng, Jianmin Bao, Dong Chen, Ming Zeng, and Fang Wen. 2021. Ex- ploring Temporal Coherence for More General Video Face Forgery Detection. In IEEE/CVF International Conference on Computer Vision (ICCV)

    2021

  55. [63]

    Dongjie Yang, Suyuan Huang, Chengqiang Lu, Xiaodong Han, Haoxin Zhang, Yan Gao, Yao Hu, and Hai Zhao. 2024. Vript: A video is worth thousands of words. Advances in Neural Information Processing Systems37 (2024), 57240–57261

    2024

  56. [64]

    Wanyi Zhuang, Qi Chu, Zhentao Tan, Qiankun Liu, Haojie Yuan, Changtao Miao, Zixiang Luo, and Nenghai Yu. 2022. UIA-ViT: Unsupervised Inconsistency-Aware Method based on Vision Transformer for Face Forgery Detection. InSpringer European Conference on Computer Vision (ECCV). A A...

    2022

  57. [67]

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. 2017. Unpaired image-to-image translation using cycle-consistent adversarial networks. InPro- ceedings of the IEEE international conference on computer vision. 2223–2232

    2017

  58. [2015]

    Imagenet large scale visual recognition challenge.International journal of computer vision115, 3 (2015), 211–252

    2015

  59. [2021]

    Laion-400m: Open dataset of clip-filtered 400 million image-text pairs.arXiv preprint arXiv:2111.02114(2021)

    2021 arXiv

  60. [2023]

    InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Learning on Gradients: Generalized Artifacts Representation for GAN- Generated Images Detection. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.