Pith. sign in

REVIEW 4 major objections 7 minor 37 references

Adapting Image-to-Video Diffusion Models for Large-Motion Frame Interpolation

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A plug-and-play conditional encoder adapts the Stable Video Diffusion image-to-video model to large-motion frame interpolation, and the paper reports the best Fréchet Video Distance on two benchmarks.

desk verdict Useful incremental extension of SVDKFI; FVD claim plausible but needs statistical hardening. read the letter →

arxiv 2412.17042 v3 pith:QXYNFK2F submitted 2024-12-22 cs.CV

Luoxu Jin , Hiroshi Watanabe This is my paper · ORCID
classification cs.CV
keywords diffusionmodelvideoframeinterpolationlargemotionconditionalencodercross-frameattentionopticalflowFréchetDistanceimage-to-videogeneration
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

The paper tries to establish that a pre-trained image-to-video diffusion model can be turned into a large-motion frame interpolator with a lightweight conditional encoder, and that this generative approach produces more realistic intermediate frames than state-of-the-art flow-based methods. The authors show this by reporting the best Fréchet Video Distance (FVD) on DAVIS-7 and UCF101-7, despite lower pixel-level metrics. A sympathetic reader would care because large and ambiguous motion is exactly where flow-based interpolation breaks down, and the proposed adaptation avoids the cost of training a video diffusion model from scratch.

What carries the argument

The central object is the conditional encoder, a copy of the Stable Video Diffusion U-Net encoder's convolutional, spatial-transformer, and temporal-transformer blocks, with zero-initialized weights and fed only clean condition inputs in the spirit of ControlNet. A dual-branch feature extractor supplies it: a spatial branch concatenates the first and last frames with zero-filled intermediate frames and a binary mask, and a temporal branch computes bidirectional optical flow via RAFT and a keyframe depth map. The branch features are reweighted with CBAM and fused with cross-attention. Inside each transformer block, a cross-frame attention mechanism uses the first and last frames (spatial block) or adjacent frames (temporal block) as key and value pairs, so every intermediate frame is reconstructed with explicit reference to the conditioning keyframes.

What would settle it

Recompute FVD on DAVIS-7 and UCF101-7 using multiple random seeds and a content-bias-corrected variant of the metric; if the advantage over AMT and FILM shrinks to noise or reverses, the paper's headline claim fails. A simpler check would be a perceptual user study on large-motion clips to see whether viewers actually prefer the diffusion-generated frames over flow-based ones.

Watch

Extended reading notes

Core claim

The central claim is that a plug-and-play conditional encoder, inserted into the pre-trained Stable Video Diffusion U-Net, turns an image-to-video generative model into a large-motion frame interpolator that produces temporally coherent intermediate frames. The encoder takes only the first and last frames as conditioned inputs: a spatial branch encodes the keyframes with a binary mask through a pre-trained autoencoder, and a temporal branch extracts bidirectional optical flow and a keyframe depth map through convolutional networks. The two branches are fused with CBAM and cross-attention, and cross-frame attention inside the spatial and temporal transformer blocks propagates keyframe appearance and temporal coherence to every query frame. The paper reports FVD of 177.79 on DAVIS-7 and 265.74 on UCF101-7, lower than AMT (279.1, 296.0), FILM (282.5, 321.2), and SVDKFI (503.0, 361.4), making it the best on distribution-level video realism, while acknowledging weaker PSNR, SSIM, LPIPS, and FID numbers.

Load-bearing premise

The central claim rests on the assumption that the Fréchet Video Distance measured on the seven generated frames is the right and reliable yardstick for interpolation quality, because on every other quantitative metric the proposed method is worse than the baselines.

Editorial extensions

If this is right

  • If the FVD result holds, generative frame interpolation can be obtained cheaply by adapting large pre-trained image-to-video models, avoiding training a video diffusion model from scratch.
  • The dual-branch design shows that explicit motion cues such as optical flow and depth still help a generative interpolator, suggesting hybrid flow-plus-diffusion architectures as a productive direction.
  • The method's ability to interpolate real-world, anime, and sketch styles indicates that the conditional encoder generalizes across visual domains from a single training run.
  • Because the model produces plausible frames even when motion is ambiguous, it could support not only slow-motion effects but creative content generation where several valid in-betweens are acceptable.
  • The reported FVD advantage over deterministic baselines, if robust, implies that distribution-level realism of interpolated video is better served by generative modeling.

Reading between the lines

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

  • The paper's loss on pixel-level metrics suggests FVD captures a distinct quality axis; a natural extension would be a forced-choice human preference test, since the paper itself references the content-bias critique of FVD in [37].
  • The cross-frame attention recipe is general and could transfer to other latent video diffusion models beyond Stable Video Diffusion, potentially improving temporal coherence for editing or inpainting tasks.
  • One testable extension is to evaluate the method with a content-bias-corrected FVD or with multiple random seeds, as the paper reports no error bars and the margin over AMT and FILM could be noisy.
  • The limitation section notes that Stable Video Diffusion struggles with ambiguous motion, so the method's gains may concentrate on large but structured motion; a dataset isolating occlusion and deformation would clarify the boundary.
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 / 7 minor

Summary. The paper proposes a conditional encoder that adapts the Stable Video Diffusion image-to-video model to video frame interpolation. The encoder uses a dual-branch feature extractor (spatial and temporal) and a cross-frame attention mechanism in the transformer blocks of the U-Net. Evaluations on DAVIS-7 and UCF101-7 report the best FVD among AMT, FILM, SVDKFI, and the proposed method, while the method is worse on PSNR, SSIM, LPIPS, and FID. The authors claim superiority on FVD, especially for large-motion scenarios, supported by one qualitative example and an ablation study.

Significance. If the FVD advantage is real and robust, the paper is a useful demonstration that pre-trained image-to-video diffusion models can be adapted for interpolation with a lightweight plug-in encoder, and the dual-branch/cross-frame attention design is a reasonable contribution. The qualitative results across real, anime, and sketch styles are appealing, and the ablation shows consistent FVD improvement from the proposed components. However, the central claim is supported only by a single distributional metric with no protocol details, no error bars, and no motion-stratified analysis, while four other metrics contradict it. The paper's own limitation section concedes weaknesses that align with the metric contradiction, so the headline result currently rests on a fragile empirical basis.

major comments (4)
  1. [Section IV-B / Table I] The FVD evaluation is under-specified. FVD is a distributional metric originally designed for 16-frame clips with I3D features, but the paper evaluates 7 generated frames and does not state the feature extractor, the number of real and generated clips, or the reference statistics. No error bars, seeds, or statistical tests are reported. Since the paper cites the content-bias critique of FVD [37] but does not address it, the reported FVD advantage over AMT and FILM could be a metric artifact. Please provide the full FVD protocol, multiple seeds or confidence intervals, and a statistical test to show the FVD gaps are meaningful.
  2. [Section IV-B / Table I] The paper loses on PSNR, SSIM, LPIPS, and FID on both datasets, yet FVD and FID share the same distribution-matching philosophy, differing mainly in temporal modeling. This pattern is exactly what one would expect from a model that produces temporally smooth but spatially less accurate frames. The citation of VIDIM [8] to dismiss the other metrics is not sufficient, especially because the paper does not provide a human study or any analysis that FVD is the decisive measure. Please add per-category or per-motion-bin FVD and FID results, or a human evaluation, to rule out a temporal-smoothing artifact.
  3. [Section IV-C / Fig. 5] The claim of superior performance 'particularly in handling large motion scenarios' is not quantitatively evaluated. There is no definition of 'large motion' (e.g., optical flow magnitude or frame gap thresholds) and no motion-stratified metrics. Only one qualitative example is shown. Add an objective motion measure and report the metrics across motion bins, or revise the claim to match the limited evidence.
  4. [Section V] The limitations section states that Stable Video Diffusion 'often struggles to generate motion-rich videos' and that the latent bottleneck compromises pixel-level metrics. These statements are in direct tension with the abstract's claim about handling large-motion scenarios. Please reconcile this tension by providing motion-specific quantitative evidence, or temper the abstract and conclusion claims accordingly.
minor comments (7)
  1. [Abstract] The sentence 'With the development of video generation models has advanced significantly in recent years' is grammatically incomplete; please rephrase.
  2. [Section III-A, Eq. (1)] The loss in Eq. (1) is written as an equality to v; it should be an L2 norm (or squared L2 norm) of the difference. Please correct the notation.
  3. [Section III-B] The sentence beginning 'As discussed in previous work [26], adding noised latents...' is a run-on; split it into two sentences for clarity.
  4. [Table II] Table II evaluates only the middle of the 7 generated frames at 512x512, whereas Table I evaluates all 7 frames. Clarify why only the center frame is used in the ablation and whether the conclusions hold for all intermediate frames.
  5. [Section IV-B / Table I] The text says 'The best results are highlighted in bold font,' but Table I contains no bold formatting. Either add bold to the best numbers or remove this sentence.
  6. [References / Section IV-B] Reference [37] is cited but never discussed; add a sentence explaining how the content-bias critique of FVD relates to the reported results.
  7. [General] No code or trained models are released, and no reproducibility details (e.g., number of denoising steps, sampling schedule, seeds) are given. Please include these or state that they are not available.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper reports an empirical comparison against external baselines with no fitted quantity renamed as a prediction and no load-bearing self-citation.

full rationale

This is an empirical systems paper. The proposed method combines a pre-trained Stable Video Diffusion backbone, a ControlNet-style conditional encoder, a dual-branch feature extractor, and cross-frame attention, trained with the standard v-prediction diffusion loss (Eq. 1). The central claim, superior FVD on DAVIS-7 and UCF101-7, is supported by direct evaluation against external baselines (AMT, FILM, SVDKFI) on public datasets. None of the components is defined in terms of FVD, and the FVD score is not optimized as a training objective nor used to select the reported configuration in a way that would make the comparison forced. The paper cites prior work for architectural inspiration (ControlNet, SparseCtrl, Stable Video Diffusion), but those citations are to external authors, not to the present authors, and they are not used to justify the empirical outcome. The discussion of metric limitations in Section IV-B and the acknowledged loss on pixel-level metrics in Section V raise questions about the strength of the FVD claim, but these are validity or robustness concerns, not circularity. No equation is equivalent to another by construction, and no fitted parameter is renamed as a prediction. Therefore the appropriate circularity score is 0.

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

The paper introduces no new physical entities or forces. Its contributions are architectural. The key assumptions are domain-level: pretrained SVD, RAFT, and MiDaS generalize to the target interpolation task, and FVD is an appropriate evaluation metric. The lack of an independent large-motion benchmark makes the FVD assumption load-bearing for the central claim.

assumptions (3)
  • domain assumption Stable Video Diffusion provides a strong prior for natural video appearance and motion that can be repurposed for frame interpolation.
    The entire method builds on SVD as the base model without modifying it (Section III-A).
  • domain assumption RAFT optical flow and MiDaS depth estimates are informative and reliable cues for interpolation.
    The temporal branch uses these pretrained extractors and feeds their outputs into the conditional encoder (Section III-B).
  • ad hoc to paper FVD is a valid primary metric for judging interpolation quality, despite the paper losing on PSNR, SSIM, LPIPS, and FID.
    The paper adopts FVD as the headline metric in Table I and Section IV-B, arguing that pixel metrics are not suitable, but does not validate this choice for the specific large-motion scenario or address the content bias warning in [37].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adapting Image-to-Video Diffusion Models for Large-Motion Frame Interpolation." pith.science (2026). https://pith.science/paper/QXYNFK2F

@misc{pith2026241217042,
  author       = {Pith},
  title        = {Pith review of: Adapting Image-to-Video Diffusion Models for Large-Motion Frame Interpolation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QXYNFK2F}},
  note         = {Machine review of arXiv:2412.17042}
}
read the original abstract

With the development of video generation models has advanced significantly in recent years, we adopt large-scale image-to-video diffusion models for video frame interpolation. We present a conditional encoder designed to adapt an image-to-video model for large-motion frame interpolation. To enhance performance, we integrate a dual-branch feature extractor and propose a cross-frame attention mechanism that effectively captures both spatial and temporal information, enabling accurate interpolations of intermediate frames. Our approach demonstrates superior performance on the Fr\'echet Video Distance (FVD) metric when evaluated against other state-of-the-art approaches, particularly in handling large motion scenarios, highlighting advancements in generative-based methodologies.

Figures

Figures reproduced from arXiv: 2412.17042 by the authors.

Figure 1
Figure 1. The input video data {I1, I2, . . . , IN } is first encoded into latent representations, where Gaussian noise is added to produce the noisy representations {z1, z2, . . . , zN }. Subsequently, the spatial and temporal features of the first and last frames are extracted through separate branches and fed into the conditional encoder. The conditional encoder integrates these features while employing zero initialization… view at source ↗
Figure 2
Figure 2. The first and last frames are processed through distinct [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. In both the spatial transformer block and the temporal [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Interpolation results across different styles. From the first row to the last row, the results correspond to real-world, [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: The example demonstrates a large-motion scenario where traditional methods fail to predict the direction and magnitude [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 29 canonical work pages

  1. [9]

    Generative inbetweening: Adapting image-to-video models for keyframe interpolation,

    X. Wang, B. Zhou, B. Curless, I. Kemelmacher-Shlizerman, A. Holyn- ski, and S. M. Seitz, “Generative inbetweening: Adapting image-to-video models for keyframe interpolation,” arXiv preprint arXiv:2408.15239 , 2024

    arXiv 2024

  2. [37]

    On the content bias in fr ´echet video distance,

    S. Ge, A. Mahapatra, G. Parmar, J.-Y . Zhu, and J.-B. Huang, “On the content bias in fr ´echet video distance,” in Computer Vision and Pattern Recognition, 2024

    work page 2024

  3. [8]

    Video interpolation with diffusion models,

    S. Jain, D. Watson, E. Tabellion, A. Holy’nski, B. Poole, and J. Kontka- nen, “Video interpolation with diffusion models,” Computer Vision and Pattern Recognition, pp. 7341–7351, 2024

    work page 2024

  4. [1]

    Video frame interpolation via deformable separable convolution,

    X. Cheng and Z. Chen, “Video frame interpolation via deformable separable convolution,” in AAAI Conference on Artificial Intelligence , 2020

    work page 2020

  5. [2]

    St-mfnet: A spatio-temporal multi- flow network for frame interpolation,

    D. Danier, F. Zhang, and D. R. Bull, “St-mfnet: A spatio-temporal multi- flow network for frame interpolation,” Computer Vision and Pattern Recognition, pp. 3511–3521, 2021

    work page 2021

  6. [3]

    Super slomo: High quality estimation of multiple intermediate frames for video interpolation,

    H. Jiang, D. Sun, V . Jampani, M.-H. Yang, E. G. Learned-Miller, and J. Kautz, “Super slomo: High quality estimation of multiple intermediate frames for video interpolation,” Computer Vision and Pattern Recogni- tion, pp. 9000–9008, 2017

    work page 2017

  7. [4]

    Softmax splatting for video frame interpolation,

    S. Niklaus and F. Liu, “Softmax splatting for video frame interpolation,” Computer Vision and Pattern Recognition , pp. 5436–5445, 2020

    work page 2020

  8. [5]

    Xvfi: extreme video frame interpolation,

    H. Sim, J. Oh, and M. Kim, “Xvfi: extreme video frame interpolation,” International Conference on Computer Vision, pp. 14 469–14 478, 2021

    work page 2021

Show all 37 references
  1. [6]

    Ifrnet: Intermediate feature refine network for efficient frame interpolation,

    L. Kong, B. Jiang, D. Luo, W. Chu, X. Huang, Y . Tai, C. Wang, and J. Yang, “Ifrnet: Intermediate feature refine network for efficient frame interpolation,” Computer Vision and Pattern Recognition , pp. 1959– 1968, 2022

    1959

  2. [7]

    Ldmvfi: Video frame interpo- lation with latent diffusion models,

    D. Danier, F. Zhang, and D. R. Bull, “Ldmvfi: Video frame interpo- lation with latent diffusion models,” in AAAI Conference on Artificial Intelligence, 2023

    2023

  3. [10]

    Framer: Interactive frame interpolation,

    W. Wang, Q. Wang, K. Zheng, H. Ouyang, Z. Chen, B. Gong, H. Chen, Y . Shen, and C. Shen, “Framer: Interactive frame interpolation,” arXiv preprint arXiv:2410.18978, 2024

    2024 arXiv

  4. [11]

    Adding conditional control to text- to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text- to-image diffusion models,” in International Conference on Computer Vision), Oct. 2023

    2023

  5. [12]

    Attention is all you need,

    A. Vaswani, N. M. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in Neural Information Processing Systems , 2017

    2017

  6. [13]

    Amt: All- pairs multi-field transforms for efficient frame interpolation,

    Z. Li, Z.-L. Zhu, L. Han, Q. Hou, C. Guo, and M.-M. Cheng, “Amt: All- pairs multi-field transforms for efficient frame interpolation,” Computer Vision and Pattern Recognition , pp. 9801–9810, 2023

    2023

  7. [14]

    Film: Frame interpolation for large motion,

    F. Reda, J. Kontkanen, E. Tabellion, D. Sun, C. Pantofaru, and B. Cur- less, “Film: Frame interpolation for large motion,” in European Confer- ence on Computer Vision . Springer, 2022, pp. 250–266

    2022

  8. [15]

    Rife: Real-time intermediate flow estimation for video frame interpolation,

    Z. Huang, T. Zhang, W. Heng, B. Shi, and S. Zhou, “Rife: Real-time intermediate flow estimation for video frame interpolation,” ArXiv, vol. abs/2011.06294, 2020

    2011 arXiv

  9. [16]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Advances in Neural Information Processing Systems , Dec. 2020

    2020

  10. [17]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in International Conference on Learning Representations , Jan. 2021

    2021

  11. [18]

    Diffusion models beat gans on image synthesis,

    P. Dhariwal and A. Nichol, “Diffusion models beat gans on image synthesis,” in Advances in Neural Information Processing Systems , Dec. 2021

    2021

  12. [19]

    Classifier-free diffusion guidance,

    J. Ho and T. Salimans, “Classifier-free diffusion guidance,” in NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applica- tions, Dec. 2021

    2021

  13. [20]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” in Computer Vision and Pattern Recognition , Jun. 2022

    2022

  14. [21]

    Stable video diffusion: Scaling latent video diffusion models to large datasets,

    A. Blattmann, T. Dockhorn, S. Kulal, D. Mendelevitch, M. Kilian, and D. Lorenz, “Stable video diffusion: Scaling latent video diffusion models to large datasets,” ArXiv, vol. abs/2311.15127, 2023

    2023 arXiv

  15. [22]

    Cogvideo: Large-scale pretraining for text-to-video generation via transformers,

    W. Hong, M. Ding, W. Zheng, X. Liu, and J. Tang, “Cogvideo: Large-scale pretraining for text-to-video generation via transformers,” in International Conference on Learning Representations , 2023

    2023

  16. [23]

    Animatediff: Animate your personalized text-to- image diffusion models without specific tuning,

    Y . Guo, C. Yang, A. Rao, Z. Liang, Y . Wang, Y . Qiao, M. Agrawala, D. Lin, and B. Dai, “Animatediff: Animate your personalized text-to- image diffusion models without specific tuning,” International Confer- ence on Learning Representations , 2024

    2024

  17. [24]

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

    A. Blattmann, R. Rombach, H. Ling, T. Dockhorn, S. W. Kim, S. Fidler, and K. Kreis, “Align your latents: High-resolution video synthesis with latent diffusion models,” Computer Vision and Pattern Recognition , pp. 22 563–22 575, 2023

    2023

  18. [25]

    Make-a- video: Text-to-video generation without text-video data,

    U. Singer, A. Polyak, T. Hayes, X. Yin, J. An, S. Zhang, Q. Hu, H. Yang, O. Ashual, O. Gafni, D. Parikh, S. Gupta, and Y . Taigman, “Make-a- video: Text-to-video generation without text-video data,” inInternational Conference on Learning Representations , 2023

    2023

  19. [26]

    Sparsectrl: Adding sparse controls to text-to-video diffusion models,

    Y . Guo, C. Yang, A. Rao, M. Agrawala, D. Lin, and B. Dai, “Sparsectrl: Adding sparse controls to text-to-video diffusion models,” ArXiv, 2023

    2023

  20. [27]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International Conference on Machine Learning , Jul. 2021

    2021

  21. [28]

    Raft: Recurrent all-pairs field transforms for optical flow,

    Z. Teed and J. Deng, “Raft: Recurrent all-pairs field transforms for optical flow,” in European Conference on Computer Vision , 2020

    2020

  22. [29]

    Towards robust monocular depth estimation: Mixing datasets for zero-shot cross- dataset transfer,

    R. Ranftl, K. Lasinger, D. Hafner, K. Schindler, and V . Koltun, “Towards robust monocular depth estimation: Mixing datasets for zero-shot cross- dataset transfer,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, pp. 1623–1637, 2019

    2019

  23. [30]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 3–19

    2018

  24. [31]

    Frozen in time: A joint video and image encoder for end-to-end retrieval,

    M. Bain, A. Nagrani, G. Varol, and A. Zisserman, “Frozen in time: A joint video and image encoder for end-to-end retrieval,” International Conference on Computer Vision , pp. 1708–1718, 2021

    2021

  25. [32]

    Video enhance- ment with task-oriented flow,

    T. Xue, B. Chen, J. Wu, D. Wei, and W. T. Freeman, “Video enhance- ment with task-oriented flow,”International Journal of Computer Vision, vol. 127, no. 8, pp. 1106–1125, 2019

    2019

  26. [33]

    The 2017 davis challenge on video object segmenta- tion,

    J. Pont-Tuset, F. Perazzi, S. Caelles, P. Arbel ´aez, A. Sorkine-Hornung, and L. V . Gool, “The 2017 davis challenge on video object segmenta- tion,” ArXiv, 2017

    2017

  27. [34]

    Ucf101: A dataset of 101 human actions classes from videos in the wild,

    K. Soomro, A. Zamir, and M. Shah, “Ucf101: A dataset of 101 human actions classes from videos in the wild,” ArXiv, 2012

    2012

  28. [35]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Computer Vision and Pattern Recognition , Jun. 2018

    2018

  29. [36]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” in Advances in Neural Information Processing Systems , Dec. 2017

    2017

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.