Pith. sign in

REVIEW 3 major objections 4 minor 40 references

This paper claims the readout, not the backbone, is the bottleneck for AI-generated video detection, and shows a 0.5M-parameter velocity-gated head reaches 95.28 AUC with the backbone frozen.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 01:18 UTC pith:JO3XFQ3Y

load-bearing objection Useful lightweight readout with strong empirical gains, but the 'readout bottleneck' claim is built on test-set diagnostics and needs a held-out design split before it is proven. the 3 major comments →

arxiv 2607.15321 v1 pith:JO3XFQ3Y submitted 2026-07-16 cs.CV

Rethinking the Readout: Unlocking Video Backbones for AI-Generated Video Detection

Manni Cui , Ziheng Qin , ZiAn Wang , Ruiqi Liu , Dianyuan Zou , Jianglan Wei , Han Zhou , Yu Liu
show 3 more authors
Jingrui Xu Wenhao Wang Zhenyu Zhang
This is my paper
classification cs.CV
keywords AI-generated video detectionreadout bottleneckpatch velocityfrozen video backbonetemporal artifactscross-generator generalizationVideoMAEvelocity-gated aggregation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Video pretrained backbones should detect AI-generated videos better than image-based detectors because the artifacts are temporal, yet in practice they often do not. This paper argues that the cause is the readout: the standard aggregation layer collapses patch-level temporal evidence into a single global vector, erasing inter-patch relations and diluting per-channel velocity magnitudes. It proposes V-PVP, a two-stream readout over patch velocities that adds about 0.5M trainable parameters, and shows that with the backbone fully frozen it reaches 95.28 AUC on AIGVDBench, exceeding end-to-end fine-tuning with the standard readout. A sympathetic reader would take the central claim to be that the readout, not the backbone, is the right place to invest capacity for AIGV detection.

Core claim

The central claim is that the readout layer, not the pretrained representation, explains why video backbones underperform on AI-generated video detection. Global pooling erases inter-patch relations and dilutes per-channel velocity magnitudes through sign cancellation. V-PVP repairs both by operating on patch velocities: a velocity-gated aggregation reweights patches by velocity magnitude, and a channel-faithful stream averages absolute per-channel velocities. With a frozen VideoMAE backbone it reaches 95.28 AUC on AIGVDBench, beats image-level and video-level baselines, transfers to GenVidBench-143k, and on four ViT backbones outperforms linear probes, matched-size heads, LoRA, and full fin

What carries the argument

The central object is the patch velocity field, the per-patch difference of frozen backbone tokens between adjacent frames. V-PVP is a readout with two parallel streams on this field: one computes softmax patch weights from an MLP over the token and its velocity magnitude, producing a velocity-gated aggregation that carries inter-patch heterogeneity into the pooled vector; the other averages absolute patch velocities per channel, preventing cancellation. A shallow temporal convolution head reads the concatenated stream over frame pairs. The design is anchored by a diagnostic: the spectral condition number of the patch-velocity matrix separates real from generated videos, and a Jensen inequal

Load-bearing premise

The load-bearing premise is that the real-vs-fake separability of patch-velocity statistics measured on the AIGVDBench test set reflects generator-invariant cues rather than benchmark-specific quirks, and that frozen patch tokens retain enough signal for readout-only learning; if either fails, V-PVP's advantage shrinks.

What would settle it

Recompute the Section 3 separability statistics on a held-out set of unseen generators and train V-PVP with the scorer conditioned only on appearance: if the real-vs-fake gap in the spectral condition number and per-channel velocity energies does not transfer, or if the appearance-only variant closes the gap on unseen generators, the central generator-invariance claim is falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Frozen video backbones can beat fully fine-tuned ones on AIGV detection when the aggregation layer is replaced, so capacity is better spent on readout design than on backbone adaptation.
  • The readout bottleneck is general across pretraining objectives: V-PVP improves TimeSformer, VideoMAE, UniFormerV2, and Qwen3-VL-8B, plus CNN backbones such as I3D-R50 and SlowFast-R50 with smaller gains.
  • V-PVP composes with backbone adaptation: LoRA and full fine-tuning add further gains (+6.8 and +8.9 AUC on AIGVDBench), while the standard readout stays nearly flat across adaptation regimes.
  • Anchoring patch selection to velocity magnitude rather than appearance limits memorization of the training generator's spatial fingerprint; alternative patch-level heads with more capacity generalize worse.
  • The paper's limitations imply boundary conditions: very short clips provide too few velocity samples, and backbones pretrained far from the target domain may not leave enough signal in frozen tokens.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: velocity-gated patch readouts may transfer to other fine-grained temporal discrimination tasks, such as video anomaly detection or audio-visual sync checking, where global pooling also suppresses local inter-patch evidence.
  • Editorial inference: the motivating statistics were computed on the same AIGVDBench test split used for evaluation; a natural follow-up would derive the readout design from training-source statistics only, or use a held-out validation set, to separate generator-invariant cues from benchmark-specific ones.
  • Editorial inference: the paper's limitations point to concrete stress tests — compression, resizing, screen recording, and re-upload may alter patch-velocity statistics more than single-frame artifacts, and generators that deliberately mimic real motion could erode the gating cue.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper argues that the principal bottleneck preventing frozen video pretrained backbones from outperforming image-level detectors on AI-generated-video (AIGV) detection is the standard readout layer, which collapses patch tokens into a single global vector. It proposes V-PVP, a lightweight readout with two parallel streams: a velocity-gated learnable aggregation (Section 4.1) and a channel-faithful mean-absolute velocity stream (Section 4.2), followed by a short temporal conv head. With VideoMAE-K400 frozen, V-PVP reports 95.28 AUC on AIGVDBench and 93.75 AUC on GenVidBench-143k, exceeding full fine-tuning under the standard readout. Ablations, cross-backbone experiments, and CNN-backbone extensions are provided.

Significance. If the causal 'readout bottleneck' claim were established, this would be a significant and practically useful result: a roughly 0.5M-parameter plug-in readout that unlocks frozen video backbones, with code released and consistent gains across benchmarks, backbones, and adaptation regimes. The paper is also commendable for evaluating on a disjoint-generator benchmark (GenVidBench-143k), reporting seed-averaged ablations, and including a CNN extension with diagnostics. However, the central explanatory claim is currently supported by a diagnostic computed on the same test split used for architecture selection and final evaluation, which materially weakens the causal interpretation. The method may still be an effective engineering contribution, but the paper's headline scientific claim is not yet established.

major comments (3)
  1. [Section 3 (Figure 2) vs Section 5.1] The design of V-PVP is motivated by statistics computed on all 63k AIGVDBench test videos (Figure 2), and the method is then evaluated on that same test split. The two design choices — velocity gating on ||v_p|| and the mean-absolute channel stream — are precisely the cues reported as separable in Figure 2. The headline 95.28 AUC therefore reflects test-set-informed architecture selection, not an independent test of the readout-bottleneck hypothesis. The GenVidBench-143k result does not fully break this circularity because its test sources (SVD, CogVideo, MuseV, Mora) overlap in generator family with AIGVDBench test sources (SVD, CogVideoX1.5, etc.). Please derive the readout using only training-generator statistics, or a pre-registered design, and then report performance on held-out test splits; also quantify how much of the 95.28 AUC survives when the architecture is fixed before seein
  2. [Appendix D.1.3 (Eqs. 13-15)] The formal derivation claims that three criteria — suppression of cancellation, sign symmetry, and first-order homogeneity — narrow the nonlinearity phi to |x|. But these criteria are introduced to yield exactly the proposed operator; no task-level argument shows they are necessary or sufficient for AIGV detection. First-order homogeneity in particular is invoked only to exclude x^2. The appendix is therefore better described as a post-hoc rationalization than a derivation. Either state a rigorous optimality criterion (e.g., a minimax variance or detection-theoretic objective that implies these conditions) or explicitly reframe Appendix D.1 as intuition-building. As written, Section 3's claim that the repair operators are formally derived is not supported.
  3. [Section 3, Eq. (3)] Loss 2 is described as the dilution of 'per-channel velocity magnitudes,' but Eq. (3) is written with the Euclidean norm ||v_t^{(p)}||_2 and Jensen's inequality for the vector norm. That inequality is not the per-channel cancellation bound that motivates r_t in Eq. (5). The correct per-channel statement appears later in Appendix D, Eq. (12). The main text should either rewrite Eq. (3) coordinate-wise or explicitly state that the per-channel bound is deferred to Appendix D. As it stands, the formalization in the main text does not match the claimed loss or the proposed repair.
minor comments (4)
  1. [Section 5.1 / Appendix B] The stopping rule 'train until loss convergence' is used for main comparisons but never defined. Report a concrete criterion (e.g., patience on validation loss or a fixed epoch count) so that the comparison is reproducible.
  2. [Figure 2] Panel (a) labels the x-axis as 'low-rank coherence (log σ1 − log σ10)' while the text calls it the spectral condition number κ_spec. Define σ1 and σ10 explicitly; a condition number would normally be σ1/σ10, so the log-difference notation needs a one-sentence explanation.
  3. [Table 3a] The row label 'ralone, MLPMean' is unclear; please spell out which head is used and clarify that 'MLPMean' is a readout-only baseline on the r stream. Also, the standard deviation of 1.84 in the 'no ||v||_2 in scorer' row is large; consider reporting per-seed results or explaining the source of variance.
  4. [Appendix F] The 'Future Works' heading appears twice, with two similar paragraphs. Merge them into one subsection.

Circularity Check

2 steps flagged

The readout-bottleneck claim rests on architecture choices selected from AIGVDBench test-set statistics and then evaluated on the same test split, making part of the reported gain circular.

specific steps
  1. fitted input called prediction [Section 3 (Figure 2) -> Section 4.1/4.2 -> Section 5.2]
    "Real vs fake statistics on all 63k AIGVDBench test videos. ... This pattern is a reliable cue for distinguishing real from fake, yet it lives in the patch dimension and disappears the moment GAP collapses it. ... V-PVP reaches mga AUC of 95.28 on AIGVDBench ... This suggests that the standard aggregation readout is an important bottleneck for video backbones in AIGV detection."

    The two V-PVP streams are direct implementations of the two statistics shown in Figure 2 to separate real from fake on the AIGVDBench test set: velocity-gated scoring conditions on ||v_p||, and the channel-faithful stream computes (1/P)Σ|v_p|. Because these operators were selected after observing their test-set separability, the 95.28 test AUC is not an independent test of the readout-bottleneck hypothesis; the architecture was fitted to the test distribution and then evaluated on it. The reported gain is therefore partly a selection artifact, not an out-of-sample confirmation of the causal claim.

  2. fitted input called prediction [Appendix D.2.1 (Table 7)]
    "For each choice we compute r_phi = 1/P Σ_p φ(v^(p)), average across time, and report the mean and standard deviation of per-channel Cohen's d across all D = 768 channels. ... The ordering | · | > ReLU > x2 ≫ id matches the prediction."

    The element-wise nonlinearity |·| is validated by measuring per-channel Cohen's d on the same AIGVDBench test set used in the final evaluation, and the final V-PVP model uses |·| in its channel-faithful stream. Since the operator was selected (or at least confirmed) by its separability on the test distribution, reporting the test AUC of the resulting model as evidence for the design is in-sample: the validation and the headline number share the same labels.

full rationale

The paper's formal derivation chain is internally coherent: the definition of patch velocity, the standard-readout family S, the Jensen bound for Loss 2, and the three criteria that narrow φ to |·| are mathematically self-contained and are not circular by themselves. The circularity is in the empirical grounding of the central claim. Figure 2 reports real-vs-fake separability of exactly the quantities later hard-coded into V-PVP (κ_spec of the patch-velocity field and per-channel mean |v_p|), computed on all 63k AIGVDBench test videos. Section 4.1 conditions the aggregation scorer on ||v_p|| to 'anchor' to that cue, and Section 4.2 adopts r_t = (1/P)Σ|v_p| because pre-pool magnitudes separate better. The headline 95.28 AUC is then reported on that same test split. This is test-set-informed architecture selection: the architecture was chosen to amplify cues discovered in the test distribution, so the test AUC is not an independent confirmation of the readout-bottleneck hypothesis. The GenVidBench-143k result partially mitigates this, but its test generators (MuseV, SVD, CogVideo, Mora) overlap in family with AIGVDBench test items (SVD, CogVideoX1.5), allowing AIGVDBench-tuned cues to transfer. No load-bearing self-citation occurs; the same-author MIRROR baseline is not central. Score 5 reflects partial circularity: the operators have independent mathematical motivation, but the central empirical claim is supported by an in-sample design-evaluation loop.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The central contribution has no new physical entities. The mathematical steps are standard (linearity, Jensen). The main burden is empirical: the paper assumes the frozen backbone's patch tokens carry the signal and that the test-set-observed velocity statistics are generator-invariant cues. The free parameters are architectural and diagnostic choices rather than fitted physical constants.

free parameters (3)
  • Projection width H = 256
    Hand-chosen width for the two-stream projections; central to keeping the readout at ~0.5M parameters.
  • Temporal convolution kernel size = 3
    Chosen to match 'the temporal scale of the inter frame inconsistencies'; justified by the authors, not by search.
  • Number of singular values in κ_spec diagnostic = 10
    The spectral condition number uses log σ1 - log σ10; the choice of 10 singular values affects the Loss 1 diagnostic, and the paper does not vary it.
axioms (5)
  • standard math Linearity of GAP and temporal difference (Eq. 2): averaging tokens then taking temporal differences equals averaging patch velocities.
    Used to characterize Loss 1 and Loss 2; follows from linearity.
  • standard math Jensen's inequality (Eq. 3 and D.1.3 Eq. 12): weighted average of magnitudes ≥ magnitude of weighted average.
    Used to show GAP dilutes per-channel velocity magnitude.
  • domain assumption Frozen patch tokens from a pretrained video backbone retain sufficient signal for readout-only detection.
    The whole method rests on this; the paper concedes in Limitations that it can fail when pretraining distribution is far from target.
  • ad hoc to paper The empirical separability of κ_spec and per-channel velocity magnitudes observed on the AIGVDBench test set is generator-invariant and holds across benchmarks.
    Used to anchor the design to a 'generator-invariant physical quantity'; only measured on the test set of the evaluation benchmark.
  • ad hoc to paper First-order homogeneity and sign symmetry are the right criteria to select the nonlinearity φ.
    Appendix D.1.3 introduces these criteria to justify |x|; they are not derived from the task and exclude alternatives like x² or smooth surrogates.

pith-pipeline@v1.3.0-alltime-deepseek · 19910 in / 14013 out tokens · 124573 ms · 2026-08-02T01:18:29.578872+00:00 · methodology

0 comments
read the original abstract

AI-generated videos (AIGVs) typically contain subtle temporal artifacts that arise from inter-frame inconsistencies rather than within individual frames. A detector that captures such artifacts should therefore benefit from video pretrained backbones over image only ones. In practice, however, video backbones with standard global readouts often fail to outperform strong image pretrained probes on AIGV benchmarks. We attribute this gap to excessive spatiotemporal aggregation in the readout. Video pretrained backbones tend to compress each frame into a single global descriptor. This compression suppresses local patch level temporal dynamics and discards inter patch relations, which are precisely the cues that AIGV detection most reliably depends on. Based on this, we propose Velocity Gated Patch Velocity Profiling (V-PVP), a lightweight readout that replaces only the aggregation layer with two parallel streams over the patch velocity field, adding only about $0.5$M trainable parameters. V-PVP serves as a general plug-and-play module that consistently improves performance across diverse video backbones under both end-to-end fine-tuning and linear probing settings. Our method reaches \textbf{95.28} AUC on AIGVDBench while keeping the backbone fully frozen. The results show that simply replacing the aggregation layer reactivates the temporal potential of frozen video backbones, restoring their advantage on AIGV detection. Code is available at https://anonymous.4open.science/r/PVP-81B3/.

Figures

Figures reproduced from arXiv: 2607.15321 by Dianyuan Zou, Han Zhou, Jianglan Wei, Jingrui Xu, Manni Cui, Ruiqi Liu, Wenhao Wang, Yu Liu, Zhenyu Zhang, Zian Wang, Ziheng Qin.

Figure 1
Figure 1. Figure 1: The readout is the bottleneck for AIGV detection. Left: AIGV detection relies on patch level temporal cues. Middle: Each colored cell represents a patch token, with color depth indicating activation strength. Standard video backbones aggregate these tokens into a single global vector via GAP or CLS, which suppresses patch level variations. Right: This collapse weakens local details and erases structured in… view at source ↗
Figure 2
Figure 2. Figure 2: The readout is the bottleneck. Real vs fake statistics on all 63k AIGVDBench test videos. 2.2 Video Backbones and Temporal Readouts Common video backbones such as TimeSformer [27], VideoMAE [28], UniFormerV2 [29], SlowFast [30], I3D [31], and XCLIP [32] are pretrained on action recognition datasets such as Kinetics-400 [33]. Downstream detection pipelines typically reuse the action-recognition readout, col… view at source ↗
Figure 3
Figure 3. Figure 3: Overall architecture of V-PVP. where z (p) t ∈ R D denotes the p-th patch token of frame t. Applying GAP across patches and then taking a temporal difference is equivalent to averaging the patch velocities, ∆¯zt = ¯zt+1 − z¯t = 1 P X P p=1 z (p) t+1 − 1 P X P p=1 z (p) t = 1 P X P p=1  z (p) t+1 − z (p) t  = 1 P X P p=1 v (p) t . (2) The patch axis collapses from P × D to 1 × D in a single step, and an a… view at source ↗
Figure 4
Figure 4. Figure 4: AUC comparison across different backbones and training strategies on [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 17 linked inside Pith

  1. [1]

    Video generation models as world simulators

    OpenAI. Video generation models as world simulators. OpenAI Sora technical report, 2024

    2024

  2. [2]

    Cogvideo: Large-scale pretraining for text-to-video generation via transformers.arXiv preprint arXiv:2205.15868, 2022

    Wenyi Hong, Ming Ding, Wendi Zheng, Xinghan Liu, and Jie Tang. Cogvideo: Large-scale pretraining for text-to-video generation via transformers.arXiv preprint arXiv:2205.15868, 2022

    Pith/arXiv arXiv 2022

  3. [3]

    Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, et al. Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024

    Pith/arXiv arXiv 2024

  4. [4]

    Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603, 2024

    Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, et al. Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603, 2024

    Pith/arXiv arXiv 2024

  5. [5]

    Open-sora: Democratizing efficient video production for all

    Zangwei Zheng, Xiangyu Peng, Tianji Yang, Chenhui Shen, Shenggui Li, Hongxin Liu, Yukun Zhou, Tianyi Li, and Yang You. Open-sora: Democratizing efficient video production for all. arXiv preprint arXiv:2412.20404, 2024

    Pith/arXiv arXiv 2024

  6. [6]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, et al. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    Pith/arXiv arXiv 2025

  7. [7]

    Demamba: Ai-generated video detection on million-scale genvideo benchmark.arXiv preprint arXiv:2405.19707, 2024

    Haoxing Chen, Yan Hong, Zizheng Huang, Zhuoer Xu, Zhangxuan Gu, Yaohui Li, Jun Lan, Huijia Zhu, Jianfu Zhang, Weiqiang Wang, et al. Demamba: Ai-generated video detection on million-scale genvideo benchmark.arXiv preprint arXiv:2405.19707, 2024

    Pith/arXiv arXiv 2024

  8. [8]

    Ai-generated video detection via spatial-temporal anomaly learning

    Jianfa Bai, Man Lin, Gang Cao, and Zijie Lou. Ai-generated video detection via spatial-temporal anomaly learning. InChinese Conference on Pattern Recognition and Computer Vision (PRCV), pages 460–470. Springer, 2024

    2024

  9. [9]

    Turns out i’m not real: Towards robust detection of ai-generated videos.arXiv preprint arXiv:2406.09601, 2024

    Qingyuan Liu, Pengyuan Shi, Yun-Yun Tsai, Chengzhi Mao, and Junfeng Yang. Turns out i’m not real: Towards robust detection of ai-generated videos.arXiv preprint arXiv:2406.09601, 2024

    Pith/arXiv arXiv 2024

  10. [10]

    Training-free detection of generated videos via spatial-temporal likelihoods.arXiv preprint arXiv:2603.15026, 2026

    Omer Ben Hayun, Roy Betser, Meir Yossef Levi, Levi Kassel, and Guy Gilboa. Training-free detection of generated videos via spatial-temporal likelihoods.arXiv preprint arXiv:2603.15026, 2026

    arXiv 2026

  11. [11]

    Rethink- ing the up-sampling operations in cnn-based generative network for generalizable deepfake detection

    Chuangchuang Tan, Yao Zhao, Shikui Wei, Guanghua Gu, Ping Liu, and Yunchao Wei. Rethink- ing the up-sampling operations in cnn-based generative network for generalizable deepfake detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 28130–28139, 2024

    2024

  12. [12]

    Dual data alignment makes ai-generated image detector easier generalizable.arXiv preprint arXiv:2505.14359, 2025

    Ruoxin Chen, Junwei Xi, Zhiyuan Yan, Ke-Yue Zhang, Shuang Wu, Jingyi Xie, Xu Chen, Lei Xu, Isabel Guan, Taiping Yao, et al. Dual data alignment makes ai-generated image detector easier generalizable.arXiv preprint arXiv:2505.14359, 2025

    arXiv 2025

  13. [13]

    Orthogonal subspace decomposition for generalizable ai-generated image detection.arXiv preprint arXiv:2411.15633, 2024

    Zhiyuan Yan, Jiangming Wang, Peng Jin, Ke-Yue Zhang, Chengchun Liu, Shen Chen, Taiping Yao, Shouhong Ding, Baoyuan Wu, and Li Yuan. Orthogonal subspace decomposition for generalizable ai-generated image detection.arXiv preprint arXiv:2411.15633, 2024

    Pith/arXiv arXiv 2024

  14. [14]

    Your one-stop solution for ai-generated video detection.arXiv preprint arXiv:2601.11035, 2026

    Long Ma, Zihao Xue, Yan Wang, Zhiyuan Yan, Jin Xu, Xiaorui Jiang, Haiyang Yu, Yong Liao, and Zhen Bi. Your one-stop solution for ai-generated video detection.arXiv preprint arXiv:2601.11035, 2026

    arXiv 2026

  15. [15]

    Physics-driven spatiotemporal modeling for ai-generated video detection

    Shuhai Zhang, ZiHao Lian, Jiahao Yang, Daiyuan Li, Guoxuan Pang, Feng Liu, Bo Han, Shutao Li, and Mingkui Tan. Physics-driven spatiotemporal modeling for ai-generated video detection. arXiv preprint arXiv:2510.08073, 2025

    Pith/arXiv arXiv 2025

  16. [16]

    Beyond spatial frequency: Pixel-wise temporal frequency-based deepfake video detection

    Taehoon Kim, Jongwook Choi, Yonghyun Jeong, Haeun Noh, Jaejun Yoo, Seungryul Baek, and Jongwon Choi. Beyond spatial frequency: Pixel-wise temporal frequency-based deepfake video detection. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11198–11207, 2025. 10

    2025

  17. [17]

    What matters in detecting ai-generated videos like sora?arXiv e-prints, pages arXiv–2406, 2024

    Chirui Chang, Zhengzhe Liu, Xiaoyang Lyu, and Xiaojuan Qi. What matters in detecting ai-generated videos like sora?arXiv e-prints, pages arXiv–2406, 2024

    2024

  18. [18]

    Detecting ai-generated video via frame consistency

    Long Ma, Zhiyuan Yan, Qinglang Guo, Yong Liao, Haiyang Yu, and Pengyuan Zhou. Detecting ai-generated video via frame consistency. In2025 IEEE International Conference on Multimedia and Expo (ICME), pages 1–6. IEEE, 2025

    2025

  19. [19]

    Skyra: Ai-generated video detection via grounded artifact reasoning.arXiv preprint arXiv:2512.15693, 2025

    Yifei Li, Wenzhao Zheng, Yanran Zhang, Runze Sun, Yu Zheng, Lei Chen, Jie Zhou, and Jiwen Lu. Skyra: Ai-generated video detection via grounded artifact reasoning.arXiv preprint arXiv:2512.15693, 2025

    Pith/arXiv arXiv 2025

  20. [20]

    Ivy-fake: A unified explainable framework and benchmark for image and video aigc detection.arXiv preprint arXiv:2506.00979, 2025

    Changjiang Jiang, Wenhui Dong, Zhonghao Zhang, Chenyang Si, Fengchang Yu, Wei Peng, Xinbin Yuan, Yifei Bi, Ming Zhao, Zian Zhou, et al. Ivy-fake: A unified explainable framework and benchmark for image and video aigc detection.arXiv preprint arXiv:2506.00979, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    Breaking latent prior bias in detectors for generalizable aigc image detection.arXiv preprint arXiv:2506.00874, 2025

    Yue Zhou, Xinan He, KaiQing Lin, Bin Fan, Feng Ding, and Bin Li. Breaking latent prior bias in detectors for generalizable aigc image detection.arXiv preprint arXiv:2506.00874, 2025

    Pith/arXiv arXiv 2025

  22. [22]

    Gc-consflow: Leveraging optical flow residuals and global context for robust deepfake detection

    Jiaxin Chen, Miao Hu, Dengyong Zhang, and Jingyang Meng. Gc-consflow: Leveraging optical flow residuals and global context for robust deepfake detection. In2025 IEEE International Conference on Multimedia and Expo (ICME), pages 1–6. IEEE, 2025

    2025

  23. [23]

    Thinking in frequency: Face forgery detection by mining frequency-aware clues

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

    2020

  24. [24]

    Cinemae: Leveraging frozen masked autoencoders for cross-generator ai image detection.arXiv preprint arXiv:2511.06325, 2025

    Minsuk Jang, Hyeonseo Jeong, Minseok Son, and Changick Kim. Cinemae: Leveraging frozen masked autoencoders for cross-generator ai image detection.arXiv preprint arXiv:2511.06325, 2025

    arXiv 2025

  25. [25]

    Towards universal fake image detectors that generalize across generative models

    Utkarsh Ojha, Yuheng Li, and Yong Jae Lee. Towards universal fake image detectors that generalize across generative models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 24480–24489, 2023

    2023

  26. [26]

    Mirror: Manifold ideal reference reconstructor for generalizable ai-generated image detection.arXiv preprint arXiv:2602.02222, 2026

    Ruiqi Liu, Manni Cui, Ziheng Qin, Zhiyuan Yan, Ruoxin Chen, Yi Han, Zhiheng Li, Junkai Chen, ZhiJin Chen, Kaiqing Lin, et al. Mirror: Manifold ideal reference reconstructor for generalizable ai-generated image detection.arXiv preprint arXiv:2602.02222, 2026

    arXiv 2026

  27. [27]

    Is space-time attention all you need for video understanding? InIcml, volume 2, page 4, 2021

    Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? InIcml, volume 2, page 4, 2021

    2021

  28. [28]

    Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training.Advances in neural information processing systems, 35:10078–10093, 2022

    Zhan Tong, Yibing Song, Jue Wang, and Limin Wang. Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training.Advances in neural information processing systems, 35:10078–10093, 2022

    2022

  29. [29]

    Uni- formerv2: Spatiotemporal learning by arming image vits with video uniformer.arXiv preprint arXiv:2211.09552, 2022

    Kunchang Li, Yali Wang, Yinan He, Yizhuo Li, Yi Wang, Limin Wang, and Yu Qiao. Uni- formerv2: Spatiotemporal learning by arming image vits with video uniformer.arXiv preprint arXiv:2211.09552, 2022

    Pith/arXiv arXiv 2022

  30. [30]

    Slowfast networks for video recognition

    Christoph Feichtenhofer, Haoqi Fan, Jitendra Malik, and Kaiming He. Slowfast networks for video recognition. InProceedings of the IEEE/CVF international conference on computer vision, pages 6202–6211, 2019

    2019

  31. [31]

    Quo vadis, action recognition? a new model and the kinetics dataset

    Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. Inproceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 6299–6308, 2017

    2017

  32. [32]

    Expanding language-image pretrained models for general video recognition

    Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, and Haibin Ling. Expanding language-image pretrained models for general video recognition. InEuropean conference on computer vision, pages 1–18. Springer, 2022

    2022

  33. [33]

    The kinetics human action video dataset.arXiv preprint arXiv:1705.06950, 2017

    Will Kay, Joao Carreira, Karen Simonyan, Brian Zhang, Chloe Hillier, Sudheendra Vijaya- narasimhan, Fabio Viola, Tim Green, Trevor Back, Paul Natsev, et al. The kinetics human action video dataset.arXiv preprint arXiv:1705.06950, 2017. 11

    Pith/arXiv arXiv 2017

  34. [34]

    Com- bining efficientnet and vision transformers for video deepfake detection

    Davide Alessandro Coccomini, Nicola Messina, Claudio Gennaro, and Fabrizio Falchi. Com- bining efficientnet and vision transformers for video deepfake detection. InInternational conference on image analysis and processing, pages 219–229. Springer, 2022

    2022

  35. [35]

    Deepfake detection scheme based on vision transformer and distillation.arXiv preprint arXiv:2104.01353, 2021

    Young-Jin Heo, Young-Ju Choi, Young-Woon Lee, and Byung-Gyu Kim. Deepfake detection scheme based on vision transformer and distillation.arXiv preprint arXiv:2104.01353, 2021

    Pith/arXiv arXiv 2021

  36. [36]

    Deepfake video detection using convolutional vision transformer.arXiv preprint arXiv:2102.11126, 2021

    Deressa Wodajo and Solomon Atnafu. Deepfake video detection using convolutional vision transformer.arXiv preprint arXiv:2102.11126, 2021

    Pith/arXiv arXiv 2021

  37. [37]

    Deep convolutional pooling trans- former for deepfake detection.ACM transactions on multimedia computing, communications and applications, 19(6):1–20, 2023

    Tianyi Wang, Harry Cheng, Kam Pui Chow, and Liqiang Nie. Deep convolutional pooling trans- former for deepfake detection.ACM transactions on multimedia computing, communications and applications, 19(6):1–20, 2023

    2023

  38. [38]

    Ai-generated video detection via perceptual straightening.arXiv preprint arXiv:2507.00583, 2025

    Christian Internò, Robert Geirhos, Markus Olhofer, Sunny Liu, Barbara Hammer, and David Klindt. Ai-generated video detection via perceptual straightening.arXiv preprint arXiv:2507.00583, 2025

    arXiv 2025

  39. [39]

    Genvidbench: A challenging benchmark for detecting ai-generated video

    Zhen-Liang Ni, Y AN Qiangyu, Tianning Yuan, Mouxiao Huang, Hailin Hu, Xinghao Chen, and Yunhe Wang. Genvidbench: A challenging benchmark for detecting ai-generated video. 2025

    2025

  40. [40]

    Qwen3-vl technical report, 2025

    Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, Mei Li, Kaixin Li, Zicheng Lin, Junyang Lin, Xuejing Liu, Jiawei Liu, Chenglong Liu, Yang Liu, Dayiheng Liu, Shixuan ...

    Pith/arXiv arXiv 2025