Pith. sign in

REVIEW 4 major objections 5 minor 6 cited by

Natural videos obey a probability-flow invariant that AI-generated videos, including Sora, measurably violate; the NSG ratio turns this violation into a reliable video detector.

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

2026-08-04 10:48 UTC pith:642QPUQD

load-bearing objection A fresh physics-based video detector with strong empirical results, but the central theorem overclaims—the upper bound doesn't establish the MMD ordering. the 4 major comments →

arxiv 2510.08073 v2 pith:642QPUQD submitted 2025-10-09 cs.CV cs.LG

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
show 1 more author
Mingkui Tan
This is my paper
classification cs.CV cs.LG
keywords AI-generated video detectiondeepfake detectionprobability flowcontinuity equationdiffusion modelsscore functionmaximum mean discrepancyspatiotemporal dynamics
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.

The paper claims that natural videos obey a probability-flow conservation law that synthetic videos systematically violate, and that this violation can be measured from pixel statistics alone. It introduces the Normalized Spatiotemporal Gradient (NSG), the ratio of the spatial gradient of log-probability to the temporal rate of density change, and estimates it using a pre-trained diffusion model's score function plus a brightness-constancy approximation of temporal derivatives. Detection then becomes a distance test: the maximum mean discrepancy between NSG features of a test video and a small reference set of real videos. The authors prove an upper bound on NSG feature distance between real and generated videos, showing the distance grows with the generator's distribution shift, and report that this detection scheme outperforms prior methods by 16.00% in recall and 10.75% in F1-score.

Core claim

The central claim is that real-video spatiotemporal dynamics can be characterized by the invariant v(x,t)·g(x,t)≈1, where v is the probability-flow velocity and g is the Normalized Spatiotemporal Gradient. Generated videos come from a shifted distribution and break this invariant, which NSG amplifies. Under Gaussian assumptions, the paper shows the squared distance between NSG features of real and generated videos is bounded by a quantity that grows with the shift φ=∥μ∥²/σ(t)², so larger distributional shifts produce larger NSG discrepancies. Because the MMD between NSG features of real videos is therefore smaller than that between real and generated videos, NSG-VD separates real from synthe

What carries the argument

The Normalized Spatiotemporal Gradient (NSG), defined as g(x,t)=∇x log p(x,t)/(−∂t log p(x,t)+λ), is the ratio that carries the argument. It is estimated without explicit motion decomposition: the spatial gradient comes from the score function sθ of a pre-trained diffusion model, and the temporal derivative comes from the brightness-constancy assumption p(x+Δx,t+Δt)≈p(x,t), which yields ∂t log p≈−∇x log p·Δx/Δt. The detection statistic is the MMD between NSG features of a test video and a reference set of real videos, with a learned deep kernel.

Load-bearing premise

The entire derivation collapses if the probability density is not approximately constant along real-video motion trajectories between consecutive frames, because then the temporal-derivative estimate no longer measures the physical quantity that NSG needs.

What would settle it

Use a physically accurate renderer with rigid-body kinematics, consistent lighting, and real sensor noise to produce synthetic videos whose NSG distances to real videos should be small if the theory is right; running NSG-VD on such videos would settle whether the detector responds to distribution shift or merely to synthetic-looking content, because if these physically correct videos are flagged as reliably as diffusion-generated videos, the claimed amplification under distribution shift is falsified.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Detectors built on the NSG invariant transfer across generation paradigms without generator-specific training, since they depend on physical constraints rather than surface artifacts.
  • The proved upper bound predicts that real-vs-fake MMD grows with distribution shift, so detection should become easier as generators drift further from real video statistics.
  • Scarce synthetic training data degrades performance gracefully: the method retains high recall on Sora and WildScrape even when trained on only 1,000 fake videos.
  • The same estimator works at reduced diffusion resolution (128x128 or 64x64), trading modest AUROC for 67 to 92 percent faster inference.
  • NSG components are complementary: spatial gradients alone give 87.99 percent recall, temporal derivatives alone 60.35 percent, while their physics-constrained ratio reaches 88.02 percent recall and 90.87 percent F1-score.

Where Pith is reading between the lines

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

  • The probability-flow invariant is not video-specific; the same ratio of spatial to temporal log-density changes could be estimated for audio, EEG, or other time-series data, potentially exposing synthetic signals by the same mechanism.
  • The brightness-constancy assumption is most plausible for slow, smooth motion; real videos with fast cuts or camera shakes may themselves produce large NSG distances, so performance on high-motion real footage should be checked before deployment.
  • Because NSG measures a physical constraint violation, it could be repurposed as a quantitative physics-faithfulness score for video generators, helping developers debug why a model's motion violates continuity.
  • The bound's dependence on φ=∥μ∥²/σ² hints at a calibration recipe: measure NSG distance on a small set of real videos to set a per-domain threshold rather than fixing a single global τ.

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

4 major / 5 minor

Summary. The paper proposes NSG-VD, a physics-inspired method for detecting AI-generated videos. The core idea is a Normalized Spatiotemporal Gradient (NSG) statistic, defined as the ratio of a spatial log-density gradient to a temporal log-density derivative, motivated by probability-flow conservation. The method estimates spatial gradients with a pre-trained diffusion model and temporal derivatives via a brightness-constancy approximation. Detection is performed by computing MMD between NSG features of a test video and a reference set of real videos, with a learned deep kernel. The authors claim a theoretical guarantee (Theorem 1) that real-vs-generated NSG feature distances are amplified by distribution shift, and they report large empirical gains on the GenVideo benchmark across multiple generators including Sora. The paper is clearly written and includes extensive ablations, efficiency analyses, and reproducibility details.

Significance. If the theoretical claim were established, this would be a useful contribution: it proposes a detector that is not tied to generator-specific artifacts and demonstrates strong cross-generator generalization, including to closed-source models such as Sora. The empirical pipeline is concrete and reproducible, with code release, parameter-efficiency analysis, ablation of the two NSG components, threshold sensitivity, and reference-set-size sensitivity. The reported gains over strong baselines are substantial and are supported by multiple tables and three-seed standard deviations. However, the central theoretical guarantee in Section 3.4 has a logical gap: the supplied upper bound does not imply the ordering between real-vs-real and real-vs-generated MMD that the method's rationale requires. The empirical results are encouraging, but the paper's headline 'proof of amplified discrepancies' is not supported by the presented mathematics.

major comments (4)
  1. [Section 3.4, Eq. (32)] Theorem 1 gives an upper bound on ||G(x)-G(y)||^2 that grows with φ, but an upper bound that grows with φ does not imply that the actual distance grows or that real-fake MMD exceeds real-real MMD. For real-vs-real pairs, φ=0 but ||G(x)-G(x')||^2 is still nonzero because of the d^2 and log^2 terms; the Gaussian kernel cross-term decreases with the actual distance, so the MMD ordering requires a lower bound on the real-fake distance (or a direct comparison of expectations), not an upper bound. The sentence 'since the distribution shift term φ=0 for real data' conflates the parameter φ with the random magnitude of NSG features. This is the load-bearing step for the abstract claim that generated videos exhibit amplified discrepancies; it needs to be fixed by deriving a lower bound or by directly bounding the MMD difference.
  2. [Section 3.2, Proposition 1 and Appendix A.2] The estimator in Eq. (7) rests on the brightness constancy assumption p(x+Δx,t+Δt)≈p(x,t), i.e., the probability density is constant along video motion trajectories. For high-dimensional, non-Gaussian video distributions this is a very strong assumption and is unlikely to hold at the pixel level, where lighting changes, occlusions, and object appearance/disappearance are common. If the assumption fails, the estimated temporal derivative no longer measures ∂_t log p, and the NSG statistic loses its physical interpretation. The paper's Appendix E.6 discusses only the divergence-term approximation, not this more fundamental assumption. The authors should either provide empirical evidence that the approximation is adequate for natural videos or explicitly reframe the method as a heuristic whose validity is demonstrated only empirically.
  3. [Section 3.4, Proposition 2] The statement of Proposition 2 contains a variance error: for x∼N(0,σ(t)^2 I_d), the quantity −x/σ(t)^2 is distributed as N(0, (1/σ(t)^2) I_d), not N(0, σ(t)^2 I_d). The same issue appears for y. While the proof of Theorem 1 uses the correct expressions in Appendix A.4, this incorrect proposition is used in the main text to motivate the theory and could mislead readers. It must be corrected.
  4. [Section 3.4 and Appendix A.7] The theorem is stated under the assumption that both real and generated videos share the same time-dependent covariance σ(t)^2 I_d and differ only by a mean shift μ. This is an acknowledged simplification, but it should be stated more prominently in the main text: the proof does not cover changes in temporal dynamics, which are the very cues the method claims to detect. The conclusion should be phrased as a heuristic motivation rather than a general guarantee.
minor comments (5)
  1. [Section 4.1] Typo: 'SENIE' should be 'SEINE' in the sentence introducing Table 2.
  2. [Appendix E.5] The text says 'These results will be included in our revision.' This is inappropriate for a submitted manuscript; the results should either be fully incorporated or omitted.
  3. [NeurIPS Checklist, item 3] The checklist says Theorem 1 provides an 'NSG feature lower bound,' but Theorem 1 is an upper bound. Please correct the wording.
  4. [Eq. (7)] The denominator is ambiguous: it should be written as sθ(xt)·(x_{t+Δt}−x_t)/Δt + λ, with explicit parentheses, to match the derivation in Eq. (9).
  5. [Figure 2] The notation in the figure is crowded and the subscript/superscript on the MMD term is unclear; consider simplifying the figure and aligning it with Eq. (10).

Circularity Check

0 steps flagged

No significant circularity: NSG is a fixed pixel-plus-score transform, the deep kernel is supervised-fit but tested on held-out generators, and the prior-work components are externally published; the flagged Theorem 1-to-MMD step is a proof gap, not a definitional reduction.

full rationale

The derivation chain is not circular by construction. (1) The NSG statistic g(x,t)=s_theta(xt)/(s_theta(xt)·(x_{t+Δt}-x_t)/Δt + λ) is a deterministic function of a frozen pre-trained diffusion score and pixel differences; no NSG parameter is fitted to the labels it is used to predict, so the representation is not defined in terms of the detection target. The brightness-constancy substitution of Proposition 1 (Eq. 9) is an explicit modeling ansatz; a subtlety is that it makes the numerator and denominator share the same score term, so the claimed dual-field relation v·g≈1 is partly enforced by the estimator's construction rather than measured from independent physics. That is a stated assumption being empirically tested, not a self-referential definition of the outcome. (2) The deep kernel and threshold are trained/tuned on real-vs-fake NSG features (Eqs. 12–13, Fig. 3), which is standard supervised fitting rather than circular reasoning: the paper's headline claims are evaluated on unseen generators (Sora, HotShot, WildScrape, Seaweed, etc.) that are external to the NSG-specific training data, so the recalls/F1 are not statistically forced by the fit. The single-sample MMD rule and MMD-MP objective are imported from the same group's prior published works ([45], [46], [48], [58]; e.g., [48] = Zhang et al., ICLR 2024, same first author), but these are released, externally checkable papers and are not the evidence for the central benchmark claims, so per the rules they count as independent support and only marginally touch the score. (3) In Section 3.4, the φ-dependence in Theorem 1's upper bound arises from a genuine computation (noncentral χ² distributions in Proposition 2, Eqs. 27–32), not from an equation defined to equal its conclusion. However, the final inference — from 'the bound grows with φ' to 'the MMD between NSGs of real videos is smaller than that between real and generated videos' — is logically unsupported: an upper bound on the real-fake distance does not imply the actual real-fake distance exceeds the real-real distance (the φ=0 bound still contains d² and log² terms, and the Gaussian-kernel MMD argument requires the opposite inequality direction, a lower bound on real-fake distance). This is a correctness gap in the theoretical guarantee as stated, not a circularity by construction. The manuscript itself acknowledges its main assumptions (incompressible-flow approximation, dependence on diffusion-model quality) in Appendix F. Overall, the c

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

No new physical entities (particles, fields, forces) are proposed. NSG is a new statistic rather than an explanatory entity, so the invented-entity ledger is empty. The main load-bearing assumptions are the incompressible-flow simplification, the probability-density brightness constancy, and the Gaussian model used for the theoretical bound.

free parameters (5)
  • lambda (NSG denominator) = 1e-10
    Stabilizes division in Eq. (6); set in Algorithm 1; hand-chosen, not central.
  • diffusion timestep for score = 5/1000
    Fixed in Appendix C.3 for all frames; affects gradient magnitude and NSG values.
  • decision threshold tau = 1.0
    Chosen from sweep in Figure 3; used in Eq. (11) for all experiments.
  • MMD deep-kernel hyperparameters = sigma_phi=0.1, sigma_Phi=100, lr=0.0001, wd=0.1
    Optimized or tuned on training real/fake videos via Eq. (13); part of the supervised detector.
  • reference set size n = 100
    Set in Appendix E.4 after sweeping; affects MMD estimate stability.
axioms (6)
  • standard math Continuity equation ∂_t p + ∇·(pv) = 0 governs any time-evolving probability density
    Used to derive Eq. (3); standard probability-flow / fluid-mechanics result [28,29].
  • domain assumption Divergence term ∇·v is subdominant (incompressible-flow approximation)
    Section 3.1, Eq. (4): needed to drop ∇·v and obtain v·g≈1; acknowledged as a simplification in Appendix E.6. If false, the physical interpretation of NSG weakens.
  • domain assumption Brightness constancy transfers to probability density: p(x+Δx,t+Δt)≈p(x,t)
    Section 3.2, Proposition 1: replaces the temporal derivative with −∇ log p·Δx/Δt; not justified for probability densities of high-dimensional video.
  • ad hoc to paper Real and generated videos are Gaussian with same covariance σ(t)^2 I_d and mean shift μ
    Section 3.4 and Appendix A; needed for Proposition 2 and Theorem 1. Real video pixel distributions are not Gaussian.
  • domain assumption Denominator bounded away from zero: |−∂_t log p + λ| ≥ C
    Theorem 1 assumption; Appendix A.7 derives conditions but C depends on the unknown ratio σ̇(t)/σ(t).
  • domain assumption Pretrained diffusion score approximates ∇_x log p of real video frames
    Section 3.2, Eq. (8): uses a guided-diffusion checkpoint. Domain shift or limited training data can degrade score estimates; the authors acknowledge this in Limitations Appendix F.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection." pith.science (2026). https://pith.science/paper/642QPUQD

@misc{pith2026251008073,
  author       = {Pith},
  title        = {Pith review of: Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/642QPUQD}},
  note         = {Machine review of arXiv:2510.08073}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

AI-generated videos have achieved near-perfect visual realism (e.g., Sora), urgently necessitating reliable detection mechanisms. However, detecting such videos faces significant challenges in modeling high-dimensional spatiotemporal dynamics and identifying subtle anomalies that violate physical laws. In this paper, we propose the first physics-driven AI-generated video detection paradigm based on probability flow conservation principles. Specifically, we propose a statistic called Normalized Spatiotemporal Gradient (NSG), which quantifies the ratio of spatial probability gradients to temporal density changes, explicitly capturing deviations from natural video dynamics. Leveraging pre-trained diffusion models, we develop an NSG estimator through spatial gradients approximation and motion-aware temporal modeling without complex motion decomposition while preserving physical constraints. Building on this, we propose an NSG-based video detection method (NSG-VD) that computes the Maximum Mean Discrepancy (MMD) between NSG features of the test and real videos as a detection metric. Last, we derive an upper bound of NSG feature distances between real and generated videos, proving that generated videos exhibit amplified discrepancies due to distributional shifts. Extensive experiments confirm that NSG-VD outperforms state-of-the-art baselines by 16.00% in Recall and 10.75% in F1-Score, validating the superior performance of NSG-VD. The source code is available at https://github.com/ZSHsh98/NSG-VD.

Figures

Figures reproduced from arXiv: 2510.08073 by Bo Han, Daiyuan Li, Feng Liu, Guoxuan Pang, Jiahao Yang, Mingkui Tan, Shuhai Zhang, Shutao Li, ZiHao Lian.

Figure 2
Figure 2. Figure 2: Overview of the proposed NSG-VD. Given a reference set of real videos [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Impact of decision threshold. We evaluate the decision threshold τ in Eqn. (11) for NSG-VD by testing τ ∈ [0.4, 1.3] under the same settings as [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparisons with baselines in terms of training costs and performance (%), where we train [PITH_FULL_IMAGE:figures/full_fig_p030_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of the values of temporal derivatives [PITH_FULL_IMAGE:figures/full_fig_p031_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Impact of reference set size for NSG-VD, where we train all models with [PITH_FULL_IMAGE:figures/full_fig_p032_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Results of the detection on real videos from the MSR-VTT dataset. Demamba NPR TALL STIL NSG-VD (Ours) Real Fake Fake Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Fake Fake Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Fake Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Fake Fake Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Fake Real Fake [PITH_FULL_IMAGE:figures/full_fig_p03… view at source ↗
Figure 8
Figure 8. Figure 8: Results of the detection on generated videos from the Crafter dataset. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Results of the detection on generated videos from the Gen2 dataset. Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake [PITH_FULL_IMAGE:figures/full_fig_p… view at source ↗
Figure 10
Figure 10. Figure 10: Results of the detection on generated videos from the HotShot dataset. 36 [PITH_FULL_IMAGE:figures/full_fig_p036_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Results of the detection on generated videos from the Lavie dataset. Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Real Fake Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Fake Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Fake Real Fake Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Fake Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Fake Real Fake Fake [PITH_FULL_IMAGE:figures/full_fig… view at source ↗
Figure 12
Figure 12. Figure 12: Results of the detection on generated videos from the ModelScope dataset. 37 [PITH_FULL_IMAGE:figures/full_fig_p037_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Results of the detection on generated videos from the MoonValley dataset. Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Real Fake Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Real Fake Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Real Fake Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Real Fake Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Real Fake Fake [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figure 14
Figure 14. Figure 14: Results of the detection on generated videos from the MorphStudio dataset. 38 [PITH_FULL_IMAGE:figures/full_fig_p038_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Results of the detection on generated videos from the Show1 dataset. Demamba NPR TALL STIL NSG-VD（Ours） Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Fake Real Fake Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours)… view at source ↗
Figure 16
Figure 16. Figure 16: Results of the detection on generated videos from the Sora dataset. 39 [PITH_FULL_IMAGE:figures/full_fig_p039_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Results of the detection on generated videos from the Seaweed dataset. Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Fake Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake Demamba NPR TALL STIL NSG-VD (Ours) Real Real Real Real Fake [PITH_FULL_IMAGE:figures/full_f… view at source ↗
Figure 18
Figure 18. Figure 18: Results of the detection on generated videos from the Seaweed dataset. 40 [PITH_FULL_IMAGE:figures/full_fig_p040_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Results of the detection on generated videos from the WildScrape dataset. To further demonstrate the excellent performance of our NSG-VD, we present visual detection results on both real and generated videos across all 10 datasets. As illustrated in Figures 7-19, both the baselines and NSG-VD demonstrate satisfactory detection on real video samples. For generated videos, the existing baselines achieve rea… view at source ↗

discussion (0)

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

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. RobustSora: De-Watermarked Benchmark for Robust AI-Generated Video Detection

    cs.CV 2025-12 conditional novelty 8.0

    RobustSora benchmark demonstrates that current AI video detectors rely heavily on visible watermarks, with average accuracy drops of 6.6 percentage points when watermarks are erased and increased false alarms when wat...

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

    cs.CV 2026-07 conditional novelty 6.0

    Replacing the global-pooling readout of a frozen video backbone with a velocity-gated, per-channel-magnitude readout improves AI-generated video detection cross-generator accuracy by several AUC points.

  3. CAM-VFD: Cross-Attention Multimodal Video Forgery Detection

    cs.CV 2026-05 unverdicted novelty 6.0

    CAM-VFD detects video forgeries by using cross-attention to identify contradictions between CLIP appearance, VideoMAE motion, and MiDaS depth features.

  4. Beyond Semantics: Uncovering the Physics of Fakes via Universal Physical Descriptors for Cross-Modal Synthetic Detection

    cs.CV 2026-04 unverdicted novelty 6.0

    Five universal physical descriptors including Laplacian variance, Sobel statistics, and residual noise variance, when integrated as text encodings with CLIP, achieve up to 99.8% accuracy detecting synthetic images acr...

  5. Skyra: AI-Generated Video Detection via Grounded Artifact Reasoning

    cs.CV 2025-12 unverdicted novelty 6.0

    Skyra is an MLLM that detects AI-generated videos by identifying and reasoning over grounded visual artifacts, supported by a new annotated dataset and benchmark.

  6. Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts

    cs.CV 2026-05 unverdicted novelty 5.0

    MDMF detects AI-generated images by learning patch-level forensic signatures and quantifying their distributional discrepancies with MMD, yielding larger separation than global methods when micro-defects are present.

Reference graph

Works this paper leans on

110 extracted references · 12 linked inside Pith · cited by 6 Pith papers

  1. [1]

    Stable video diffusion: Scaling latent video diffusion models to large datasets.arXiv preprint arXiv:2311.15127, 2023

    Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Do- minik Lorenz, Yam Levi, Zion English, Vikram V oleti, Adam Letts, et al. Stable video diffusion: Scaling latent video diffusion models to large datasets.arXiv preprint arXiv:2311.15127, 2023

    Pith/arXiv arXiv 2023

  2. [2]

    Align your latents: High-resolution video synthesis with latent diffusion models

    Andreas Blattmann, Robin Rombach, Huan Ling, Tim Dockhorn, Seung Wook Kim, Sanja Fidler, and Karsten Kreis. Align your latents: High-resolution video synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22563–22575, 2023

    2023

  3. [3]

    Video generation models as world simulators

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luhman, Eric Luhman, et al. Video generation models as world simulators. OpenAI Blog, 1:8, 2024

    2024

  4. [4]

    Customcrafter: Customized video generation with preserving motion and concept composition abilities

    Tao Wu, Yong Zhang, Xintao Wang, Xianpan Zhou, Guangcong Zheng, Zhongang Qi, Ying Shan, and Xi Li. Customcrafter: Customized video generation with preserving motion and concept composition abilities. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 8469–8477, 2025

    2025

  5. [5]

    Unveiling causal reasoning in large language models: Reality or mirage? 2024

    Haoang Chi, He Li, Wenjing Yang, Feng Liu, Long Lan, Xiaoguang Ren, Tongliang Liu, and Bo Han. Unveiling causal reasoning in large language models: Reality or mirage? 2024. 10

    2024

  6. [6]

    Fine-grained controllable video generation via object appearance and context

    Hsin-Ping Huang, Yu-Chuan Su, Deqing Sun, Lu Jiang, Xuhui Jia, Yukun Zhu, and Ming- Hsuan Yang. Fine-grained controllable video generation via object appearance and context. In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3698–3708. IEEE, 2025

    2025

  7. [7]

    Videocrafter2: Overcoming data limitations for high-quality video diffusion models

    Haoxin Chen, Yong Zhang, Xiaodong Cun, Menghan Xia, Xintao Wang, Chao Weng, and Ying Shan. Videocrafter2: Overcoming data limitations for high-quality video diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7310–7320, 2024

    2024

  8. [8]

    Dynamicrafter: Animating open-domain images with video diffusion priors

    Jinbo Xing, Menghan Xia, Yong Zhang, Haoxin Chen, Wangbo Yu, Hanyuan Liu, Gongye Liu, Xintao Wang, Ying Shan, and Tien-Tsin Wong. Dynamicrafter: Animating open-domain images with video diffusion priors. InEuropean Conference on Computer Vision, pages 399–417. Springer, 2024

    2024

  9. [9]

    Motion-i2v: Consistent and controllable image-to-video generation with explicit motion modeling

    Xiaoyu Shi, Zhaoyang Huang, Fu-Yun Wang, Weikang Bian, Dasong Li, Yi Zhang, Manyuan Zhang, Ka Chun Cheung, Simon See, Hongwei Qin, et al. Motion-i2v: Consistent and controllable image-to-video generation with explicit motion modeling. InACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024

    2024

  10. [10]

    Image conductor: Precision control for interactive video synthesis

    Yaowei Li, Xintao Wang, Zhaoyang Zhang, Zhouxia Wang, Ziyang Yuan, Liangbin Xie, Ying Shan, and Yuexian Zou. Image conductor: Precision control for interactive video synthesis. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 5031–5038, 2025

    2025

  11. [11]

    Towards open-set identity preserving face synthesis

    Jianmin Bao, Dong Chen, Fang Wen, Houqiang Li, and Gang Hua. Towards open-set identity preserving face synthesis. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 6713–6722, 2018

    2018

  12. [12]

    Pulid: Pure and lightning id customization via contrastive alignment.Advances in neural information processing systems, 37:36777–36804, 2024

    Zinan Guo, Yanze Wu, Chen Zhuowei, Peng Zhang, Qian He, et al. Pulid: Pure and lightning id customization via contrastive alignment.Advances in neural information processing systems, 37:36777–36804, 2024

    2024

  13. [13]

    Out-of-distribution detection learning with unreliable out-of-distribution sources

    Haotian Zheng, Qizhou Wang, Zhen Fang, Xiaobo Xia, Feng Liu, Tongliang Liu, and Bo Han. Out-of-distribution detection learning with unreliable out-of-distribution sources. InNeurIPS, 2023

    2023

  14. [14]

    Hififace: 3d shape and semantic prior guided high fidelity face swapping

    Yuhan Wang, Xu Chen, Junwei Zhu, Wenqing Chu, Ying Tai, Chengjie Wang, Jilin Li, Yongjian Wu, Feiyue Huang, and Rongrong Ji. Hififace: 3d shape and semantic prior guided high fidelity face swapping. InProceedings of the Thirtieth International Joint Conference on Artificial Intelligence, pages 1136–1142. International Joint Conferences on Artificial Intel...

    2021

  15. [15]

    Diffswap: High-fidelity and controllable face swapping via 3d-aware masked diffusion

    Wenliang Zhao, Yongming Rao, Weikang Shi, Zuyan Liu, Jie Zhou, and Jiwen Lu. Diffswap: High-fidelity and controllable face swapping via 3d-aware masked diffusion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8568–8577, 2023

    2023

  16. [16]

    Avff: Audio-visual feature fusion for video deepfake detection

    Trevine Oorloff, Surya Koppisetti, Nicolò Bonettini, Divyaraj Solanki, Ben Colman, Yaser Yacoob, Ali Shahriyari, and Gaurav Bharaj. Avff: Audio-visual feature fusion for video deepfake detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27102–27112, 2024

    2024

  17. [17]

    Learning defense transformations for counterattacking adversarial examples.Neural Networks, 164:177–185, 2023

    Jincheng Li, Shuhai Zhang, Jiezhang Cao, and Mingkui Tan. Learning defense transformations for counterattacking adversarial examples.Neural Networks, 164:177–185, 2023

    2023

  18. [18]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  19. [19]

    Vbench: Comprehensive benchmark suite for video 11 generative models

    Ziqi Huang, Yinan He, Jiashuo Yu, Fan Zhang, Chenyang Si, Yuming Jiang, Yuanhan Zhang, Tianxing Wu, Qingyang Jin, Nattapol Chanpaisit, Yaohui Wang, Xinyuan Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. Vbench: Comprehensive benchmark suite for video 11 generative models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogn...

    2024

  20. [20]

    Vbench-2.0: Advancing video generation benchmark suite for intrinsic faithfulness.arXiv preprint arXiv:2503.21755, 2025

    Dian Zheng, Ziqi Huang, Hongbo Liu, Kai Zou, Yinan He, Fan Zhang, Yuanhan Zhang, Jingwen He, Wei-Shi Zheng, Yu Qiao, et al. Vbench-2.0: Advancing video generation benchmark suite for intrinsic faithfulness.arXiv preprint arXiv:2503.21755, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    Codef: Content deformation fields for temporally consistent video processing

    Hao Ouyang, Qiuyu Wang, Yuxi Xiao, Qingyan Bai, Juntao Zhang, Kecheng Zheng, Xiaowei Zhou, Qifeng Chen, and Yujun Shen. Codef: Content deformation fields for temporally consistent video processing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8089–8099, 2024

    2024

  22. [22]

    Deepfake video detection through optical flow based cnn

    Irene Amerini, Leonardo Galteri, Roberto Caldelli, and Alberto Del Bimbo. Deepfake video detection through optical flow based cnn. InProceedings of the IEEE/CVF international conference on computer vision workshops, pages 0–0, 2019

    2019

  23. [23]

    Altfreezing for more general video face forgery detection

    Zhendong Wang, Jianmin Bao, Wengang Zhou, Weilun Wang, and Houqiang Li. Altfreezing for more general video face forgery detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4129–4138, 2023

    2023

  24. [24]

    Tall: Thumb- nail layout for deepfake video detection

    Yuting Xu, Jian Liang, Gengyun Jia, Ziming Yang, Yanhao Zhang, and Ran He. Tall: Thumb- nail layout for deepfake video detection. InProceedings of the IEEE/CVF international conference on computer vision, pages 22658–22668, 2023

    2023

  25. [25]

    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

  26. [26]

    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

  27. [27]

    Rethinking 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. Rethinking the up-sampling operations in cnn-based generative network for generalizable deepfake detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 28130–28139, 2024

    2024

  28. [28]

    Cambridge university press, 2000

    George Keith Batchelor.An introduction to fluid dynamics. Cambridge university press, 2000

    2000

  29. [29]

    Springer, 2014

    Michel Rieutord.Fluid dynamics: an introduction. Springer, 2014

    2014

  30. [30]

    Generative modeling by estimating gradients of the data distribution.Advances in Neural Information Processing Systems, 32, 2019

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution.Advances in Neural Information Processing Systems, 32, 2019

    2019

  31. [31]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021

    2021

  32. [32]

    Determining optical flow.Artificial intelligence, 17(1-3):185–203, 1981

    Berthold KP Horn and Brian G Schunck. Determining optical flow.Artificial intelligence, 17(1-3):185–203, 1981

    1981

  33. [33]

    A kernel two-sample test.Journal of Machine Learning Research, 13(1):723–773, 2012

    Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Schölkopf, and Alexander Smola. A kernel two-sample test.Journal of Machine Learning Research, 13(1):723–773, 2012

    2012

  34. [34]

    Learning deep kernels for non-parametric two-sample tests

    Feng Liu, Wenkai Xu, Jie Lu, Guangquan Zhang, Arthur Gretton, and Danica J Sutherland. Learning deep kernels for non-parametric two-sample tests. InInternational Conference on Machine Learning, pages 6316–6326. PMLR, 2020

    2020

  35. [35]

    Exposing deep fakes using inconsistent head poses

    Xin Yang, Yuezun Li, and Siwei Lyu. Exposing deep fakes using inconsistent head poses. In ICASSP 2019-2019 IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 8261–8265. IEEE, 2019. 12

    2019

  36. [36]

    Spatiotemporal inconsistency learning for deepfake video detection

    Zhihao Gu, Yang Chen, Taiping Yao, Shouhong Ding, Jilin Li, Feiyue Huang, and Lizhuang Ma. Spatiotemporal inconsistency learning for deepfake video detection. InProceedings of the 29th ACM international conference on multimedia, pages 3473–3481, 2021

    2021

  37. [37]

    Where deepfakes gaze at? spatial-temporal gaze inconsistency analysis for video face forgery detec- tion.IEEE Transactions on Information Forensics and Security, 2024

    Chunlei Peng, Zimin Miao, Decheng Liu, Nannan Wang, Ruimin Hu, and Xinbo Gao. Where deepfakes gaze at? spatial-temporal gaze inconsistency analysis for video face forgery detec- tion.IEEE Transactions on Information Forensics and Security, 2024

    2024

  38. [38]

    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

  39. [39]

    Decof: Generated video detection via frame consistency: The first benchmark dataset.arXiv preprint arXiv:2402.02085, 2024

    Long Ma, Jiajia Zhang, Hongping Deng, Ningyu Zhang, Qinglang Guo, Haiyang Yu, Yong Liao, and Pengyuan Zhou. Decof: Generated video detection via frame consistency: The first benchmark dataset.arXiv preprint arXiv:2402.02085, 2024

    Pith/arXiv arXiv 2024

  40. [40]

    On learning multi-modal forgery representation for diffusion generated video detection

    Xiufeng Song, Xiao Guo, Jiache Zhang, Qirui Li, LEI BAI, Xiaoming Liu, Guangtao Zhai, and Xiaohong Liu. On learning multi-modal forgery representation for diffusion generated video detection. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems

  41. [41]

    Denoising diffusion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    2020

  42. [42]

    Improved techniques for training score-based generative models

    Yang Song and Stefano Ermon. Improved techniques for training score-based generative models. InNeurIPS, volume 33, pages 12438–12448, 2020

    2020

  43. [43]

    Diffusion models for adversarial purification

    Weili Nie, Brandon Guo, Yujia Huang, Chaowei Xiao, Arash Vahdat, and Animashree Anand- kumar. Diffusion models for adversarial purification. InICML, pages 16805–16827. PMLR, 2022

    2022

  44. [44]

    Adversarial purification with score-based generative models

    Jongmin Yoon, Sung Ju Hwang, and Juho Lee. Adversarial purification with score-based generative models. InICML, pages 12062–12072. PMLR, 2021

    2021

  45. [45]

    Detecting adversarial data by probing multiple perturbations using expected perturbation score

    Shuhai Zhang, Feng Liu, Jiahao Yang, Yifan Yang, Changsheng Li, Bo Han, and Mingkui Tan. Detecting adversarial data by probing multiple perturbations using expected perturbation score. InInternational Conference on Machine Learning, pages 41429–41451. PMLR, 2023

    2023

  46. [46]

    Deep kernel relative test for machine-generated text detection

    Yiliao Song, Zhenqiao Yuan, Shuhai Zhang, Zhen Fang, Jun Yu, and Feng Liu. Deep kernel relative test for machine-generated text detection. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  47. [47]

    Dire for diffusion-generated image detection

    Zhendong Wang, Jianmin Bao, Wengang Zhou, Weilun Wang, Hezhen Hu, Hong Chen, and Houqiang Li. Dire for diffusion-generated image detection. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 22445–22455, 2023

    2023

  48. [48]

    Detect- ing machine-generated texts by multi-population aware optimization for maximum mean discrepancy

    Shuhai Zhang, Yiliao Song, Jiahao Yang, Yuanqing Li, Bo Han, and Mingkui Tan. Detect- ing machine-generated texts by multi-population aware optimization for maximum mean discrepancy. InInternational Conference on Learning Representations, 2024

    2024

  49. [49]

    Generating long videos of dynamic scenes

    Tim Brooks, Janne Hellsten, Miika Aittala, Ting-Chun Wang, Timo Aila, Jaakko Lehtinen, Ming-Yu Liu, Alexei Efros, and Tero Karras. Generating long videos of dynamic scenes. Advances in Neural Information Processing Systems, 35:31769–31781, 2022

    2022

  50. [50]

    Quantum field theory.Reviews of Modern Physics, 71(2):S85, 1999

    Frank Wilczek. Quantum field theory.Reviews of Modern Physics, 71(2):S85, 1999

    1999

  51. [51]

    Courier Corporation, 1978

    John Lighton Synge and Alfred Schild.Tensor calculus, volume 5. Courier Corporation, 1978

    1978

  52. [52]

    Electron spin and probability current density in quantum mechanics.American Journal of Physics, 82(7):681–690, 2014

    WB Hodge, SV Migirditch, and William C Kerr. Electron spin and probability current density in quantum mechanics.American Journal of Physics, 82(7):681–690, 2014

    2014

  53. [53]

    Fokker-planck equation

    Hannes Risken. Fokker-planck equation. InThe Fokker-Planck equation: methods of solution and applications, pages 63–95. Springer, 1989

    1989

  54. [54]

    John Wiley & Sons, 2024

    Ronald L Panton.Incompressible flow. John Wiley & Sons, 2024. 13

    2024

  55. [55]

    Springer Science & Business Media, 2013

    Arno Böhm.Quantum mechanics: foundations and applications. Springer Science & Business Media, 2013

    2013

  56. [56]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in Neural Information Processing Systems, 34:8780–8794, 2021

    2021

  57. [57]

    Maximum mean discrepancy test is aware of adversarial attacks

    Ruize Gao, Feng Liu, Jingfeng Zhang, Bo Han, Tongliang Liu, Gang Niu, and Masashi Sugiyama. Maximum mean discrepancy test is aware of adversarial attacks. InInternational Conference on Machine Learning, pages 3564–3575. PMLR, 2021

    2021

  58. [58]

    Detecting adversarial data by probing multiple perturbations using expected perturbation score

    Shuhai Zhang, Feng Liu, Jiahao Yang, Yifan Yang, Changsheng Li, Bo Han, and Mingkui Tan. Detecting adversarial data by probing multiple perturbations using expected perturbation score. InInternational conference on machine learning, pages 41429–41451. PMLR, 2023

    2023

  59. [59]

    Toward understanding generative data augmentation.Advances in neural information processing systems, 36:54046–54060, 2023

    Chenyu Zheng, Guoqiang Wu, and Chongxuan Li. Toward understanding generative data augmentation.Advances in neural information processing systems, 36:54046–54060, 2023

    2023

  60. [60]

    Embedding trajectory for out-of-distribution detection in mathematical reasoning.Advances in Neural Information Processing Systems, 37:42965–42999, 2024

    Yiming Wang, Pei Zhang, Baosong Yang, Derek Wong, Zhuosheng Zhang, and Rui Wang. Embedding trajectory for out-of-distribution detection in mathematical reasoning.Advances in Neural Information Processing Systems, 37:42965–42999, 2024

    2024

  61. [61]

    Approximate formulae for the percentage points and the probability integral of the non-centralχ2 distribution.Biometrika, 41(3/4):538–540, 1954

    SH Abdel-Aty. Approximate formulae for the percentage points and the probability integral of the non-centralχ2 distribution.Biometrika, 41(3/4):538–540, 1954

    1954

  62. [62]

    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

    Pith/arXiv arXiv 2017

  63. [63]

    Seine: Short-to-long video diffusion model for generative transition and prediction

    Xinyuan Chen, Yaohui Wang, Lingjun Zhang, Shaobin Zhuang, Xin Ma, Jiashuo Yu, Yali Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. Seine: Short-to-long video diffusion model for generative transition and prediction. InThe Twelfth International Conference on Learning Representations, 2023

    2023

  64. [64]

    Pika: Empowering non-programmers to author executable governance policies in online communities

    Leijie Wang, Nicholas Vincent, Julija Rukanskait˙e, and Amy Xian Zhang. Pika: Empowering non-programmers to author executable governance policies in online communities. InPro- ceedings of the 2024 CHI Conference on Human Factors in Computing Systems, pages 1–18, 2024

    2024

  65. [65]

    Msr-vtt: A large video description dataset for bridging video and language

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

    2016

  66. [66]

    Muc-5 evaluation metrics

    Nancy Chinchor and Beth M Sundheim. Muc-5 evaluation metrics. InFifth Message Under- standing Conference (MUC-5): Proceedings of a Conference Held in Baltimore, Maryland, August 25-27, 1993, 1993

    1993

  67. [67]

    Using auc and accuracy in evaluating learning algorithms

    Jin Huang and Charles X Ling. Using auc and accuracy in evaluating learning algorithms. IEEE Transactions on knowledge and Data Engineering, 17(3):299–310, 2005

    2005

  68. [68]

    Minimum contrast estimators on sieves: exponential bounds and rates of convergence

    Lucien Birgé and Pascal Massart. Minimum contrast estimators on sieves: exponential bounds and rates of convergence. 1998

    1998

  69. [69]

    Kernel-based tests for likelihood-free hypothesis testing.Advances in Neural Information Processing Systems, 36:15680–15715, 2023

    Patrik Róbert Gerber, Tianze Jiang, Yury Polyanskiy, and Rui Sun. Kernel-based tests for likelihood-free hypothesis testing.Advances in Neural Information Processing Systems, 36:15680–15715, 2023

    2023

  70. [70]

    Nyström m-hilbert-schmidt independence criterion

    Florian Kalinke and Zoltán Szabó. Nyström m-hilbert-schmidt independence criterion. In Uncertainty in Artificial Intelligence, pages 1005–1015. PMLR, 2023

    2023

  71. [71]

    Trustworthy machine learning: From data to models.Foundations and Trends® in Privacy and Security, 7(2-3):74–246, 2025

    Bo Han, Jiangchao Yao, Tongliang Liu, Bo Li, Sanmi Koyejo, Feng Liu, et al. Trustworthy machine learning: From data to models.Foundations and Trends® in Privacy and Security, 7(2-3):74–246, 2025

    2025

  72. [72]

    Integral probability metrics and their generating classes of functions.Advances in applied probability, 29(2):429–443, 1997

    Alfred Müller. Integral probability metrics and their generating classes of functions.Advances in applied probability, 29(2):429–443, 1997. 14

    1997

  73. [73]

    Cheung, and James T

    Haoang Chi, Feng Liu, Wenjing Yang, Long Lan, Tongliang Liu, Bo Han, William K. Cheung, and James T. Kwok. TOHAN: A one-step approach towards few-shot hypothesis adaptation. InNeurIPS, 2021

    2021

  74. [74]

    Minimax estimation of maximum mean discrepancy with radial kernels.Advances in Neural Information Processing Systems, 29, 2016

    Ilya O Tolstikhin, Bharath K Sriperumbudur, and Bernhard Schölkopf. Minimax estimation of maximum mean discrepancy with radial kernels.Advances in Neural Information Processing Systems, 29, 2016

    2016

  75. [75]

    Minimax optimality of permuta- tion tests.The Annals of Statistics, 50(1):225–251, 2022

    Ilmun Kim, Sivaraman Balakrishnan, and Larry Wasserman. Minimax optimality of permuta- tion tests.The Annals of Statistics, 50(1):225–251, 2022

    2022

  76. [76]

    Youku-mplug: A 10 million large-scale chinese video-language dataset for pre-training and benchmarks.arXiv preprint arXiv:2306.04362, 2023

    Haiyang Xu, Qinghao Ye, Xuan Wu, Ming Yan, Yuan Miao, Jiabo Ye, Guohai Xu, Anwen Hu, Yaya Shi, Guangwei Xu, et al. Youku-mplug: A 10 million large-scale chinese video-language dataset for pre-training and benchmarks.arXiv preprint arXiv:2306.04362, 2023

    Pith/arXiv arXiv 2023

  77. [77]

    Zeroscope. 2023. URLhttps://huggingface.co/cerspense/zeroscope_v2_XL

    2023

  78. [78]

    I2vgen-xl: High-quality image-to-video synthesis via cascaded diffusion models.arXiv preprint arXiv:2311.04145, 2023

    Shiwei Zhang, Jiayu Wang, Yingya Zhang, Kang Zhao, Hangjie Yuan, Zhiwu Qin, Xiang Wang, Deli Zhao, and Jingren Zhou. I2vgen-xl: High-quality image-to-video synthesis via cascaded diffusion models.arXiv preprint arXiv:2311.04145, 2023

    Pith/arXiv arXiv 2023

  79. [79]

    Pia: Your person- alized image animator via plug-and-play modules in text-to-image models

    Yiming Zhang, Zhening Xing, Yanhong Zeng, Youqing Fang, and Kai Chen. Pia: Your person- alized image animator via plug-and-play modules in text-to-image models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7747–7756, 2024

    2024

  80. [80]

    Latte: Latent diffusion transformer for video generation.arXiv preprint arXiv:2401.03048, 2024

    Xin Ma, Yaohui Wang, Gengyun Jia, Xinyuan Chen, Ziwei Liu, Yuan-Fang Li, Cunjian Chen, and Yu Qiao. Latte: Latent diffusion transformer for video generation.arXiv preprint arXiv:2401.03048, 2024

    Pith/arXiv arXiv 2024

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.