Pith. sign in

REVIEW 4 major objections 7 minor 35 references

Learning Variations in Human Motion via Mix-and-Match Perturbation

T0 review · 4 major / 7 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper claims that randomly mixing noise into an RNN's hidden state forces stochastic motion prediction models to produce genuinely diverse, high-quality futures.

desk verdict A simple, useful trick for preventing latent-code collapse in conditional VAEs, but the quality half of the claim needs a stronger defense than a small human study. read the letter →

arxiv 1908.00733 v2 pith:WUOGGNMV submitted 2019-08-02 cs.LG cs.CVstat.ML

Mohammad Sadegh Aliakbarian , Fatemeh Sadat Saleh , Mathieu Salzmann , Lars Petersson , Stephen Gould , Amirhossein Habibian This is my paper · ORCID
classification cs.LGcs.CVstat.ML
keywords humanmotionpredictionstochasticgenerationconditionalvariationalautoencodermix-and-matchperturbationdiversityvsqualitytrade-offrecurrentencoder-decodercurriculumlearningquaternionposerepresentation
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 takes on a failure mode in stochastic human-motion prediction: neural nets trained to forecast several plausible futures tend to quietly ignore the injected randomness and collapse to one deterministic prediction. It proposes Mix-and-Match perturbation, which randomly selects which coordinates of the RNN hidden state are replaced by noise at every training step, so the model cannot learn to discount the noise. The authors report that this keeps decoder-input diversity growing through training, yields motions judged realistic by both a classifier and human raters, and matches deterministic long-term prediction accuracy when the best of K sampled futures is scored. If correct, the work shows that the way randomness is fused into a conditional generator, not just the amount, determines whether the model produces genuinely multiple futures.

What carries the argument

The load-bearing object is the Mix-and-Match perturbation operator. Given a hidden state $h_t$ and a noise vector $z$, it samples a set $I$ of $\lceil \alpha L \rceil$ indices uniformly at random, places the hidden state's values at $I$ into the output, fills the complementary indices with $z$, and then runs the mixed vector through a small residual block before fusing it with the untouched part of $h_t$. The random index set is redrawn for every mini-batch (and shared across the perturbations inside one batch), so no fixed subspace of the hidden state is reserved for the noise. A curriculum schedule—starting with a deterministic mask and gradually randomizing the chosen indices—stabilizes training when $\alpha$ is large. This operator does the work of forcing the conditional VAE's decoder to rely on $z$, converting a latent code the network could learn to ignore into one it must use.

What would settle it

Train the same architecture on motion samples whose ground-truth variation is concentrated in a few hidden-state dimensions, such as one limb driving all differences between futures, then run Mix-and-Match with masks that always exclude those dimensions; if predictions lose diversity or quality in that setting, the redundancy assumption is falsified. A simpler check is to take a trained model and decode the same observation with two masks that share no indices; if the outputs are statistically indistinguishable across many samples, the perturbation is not actually carrying the variation.

Watch

Extended reading notes

Core claim

The central claim is that prior stochastic motion-prediction models—whether they concatenate noise to a hidden state, add a transformed noise vector, or perturb the pose—give the network enough flexibility to ignore the source of variation, so their outputs become nearly identical. Mix-and-Match perturbation instead replaces a randomly chosen subset of hidden-state entries with standard normal noise, making the noise spatially interleaved with the conditioning signal and re-sampling a new random mask every mini-batch. The decoder therefore cannot route around the noise, and the paper's experiments show the diversity of the decoder inputs increases with training rather than collapsing, while the generated motions retain high quality. The authors further claim that, measured by best-of-K error on long-term prediction, this diversity costs nothing in accuracy: the model matches deterministic state-of-the-art methods while also supplying many plausible futures, and does so without requiring action labels.

Load-bearing premise

The method assumes the RNN hidden state is redundant enough that a random subset of its entries, with the rest replaced by Gaussian noise, still leaves the decoder enough motion information to produce realistic poses; if the signal for future motion is concentrated in a few dimensions, random masking will discard it and quality will drop.

Editorial extensions

If this is right

  • A conditional RNN decoder trained with Mix-and-Match keeps the diversity of its latent inputs rising throughout training, while concatenation- and addition-based baselines see diversity fall as the network learns to ignore noise.
  • The model produces $K=50$ futures whose best-of-$K$ mean angle error matches deterministic state-of-the-art predictors up to 1000 ms, so diversity is not obtained by sacrificing worst-case accuracy.
  • Quality and diversity vary monotonically with the single hyperparameter $\alpha$; values in $[0.3, 0.7]$ yield both high judged realism and high diversity, giving a task-dependent dial rather than a knife-edge setting.
  • The same architecture needs no action labels at training or inference, unlike several deterministic baselines, and can be applied action-agnostically to arbitrary observed motion.

Reading between the lines

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

  • The same masking trick should transfer to other conditional generative sequence tasks—text, speech, control—where the conditioning signal dominates and models are known to collapse to deterministic outputs; the paper identifies the mechanism generically but tests it only on motion.
  • Because the mask is redrawn per mini-batch and shared across all time steps, the model effectively sees a different latent subspace per sample; one could test whether the chosen mask correlates with interpretable variation, such as which body part deviates, which would make the latent code more controllable.
  • A direct ablation to run is to keep the mask fixed to a small set of high-variance hidden dimensions; if quality stays high but diversity collapses, the redundancy assumption is confirmed, whereas if both stay high the method may work through a different mechanism such as implicit regularization.
Share X Bluesky LinkedIn Reddit HN

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 Mix-and-Match perturbation for stochastic human motion prediction. Instead of concatenating a noise vector with the conditioning hidden state in a deterministic way, the method randomly selects a subset of hidden-state entries and replaces them with samples from a standard normal distribution, forcing the network to use the noise. This is embedded in a recurrent encoder-decoder with a CVAE block, and a curriculum schedule gradually randomizes the selected indices during training. The paper introduces two quantitative metrics, a classifier-based quality metric and an average-pairwise-distance diversity metric, and reports experiments on Human3.6M showing that the proposed model generates more diverse motions than the stochastic baselines LHP, RHP, and LPP while maintaining comparable or superior quality, along with a comparison to deterministic state-of-the-art methods using the best-of-K criterion.

Significance. If the claims hold, the paper makes a useful contribution by addressing a known failure mode of conditional generative models: the tendency to ignore the latent noise source. The proposed Mix-and-Match perturbation is simple and plausible, and the paper provides an empirical demonstration that existing methods' hidden-state diversity collapses during training. The experiments are conducted with consistent backbone architectures for the baselines, and the evaluation includes both automatic metrics and human judgment. The paper also introduces a falsifiable prediction, namely that perturbing randomly selected hidden units preserves both diversity and realism. However, the evidence for the 'high-quality' half of the central claim is weaker than the evidence for diversity, and the loss equations contain sign inconsistencies that need to be resolved.

major comments (4)
  1. [Section 3.2, Eqs. (2)-(5)] The loss terms Lrot and Lskl are defined with negative signs in Eqs. (3) and (4), and Lprior is defined as a negative KL divergence in Eq. (5). Because the text says the model is trained by minimizing the total loss in Eq. (2), these signs imply that minimizing the reconstruction loss would maximize the mean squared error, and that the KL term would be subtracted rather than added as in the standard CVAE objective. This is inconsistent with standard practice and with the reported successful results; the authors need to correct the equations or explicitly state that a maximization is intended.
  2. [Section 4.1, Figs. 6 and 7] The classifier-based quality metric shows the proposed method underperforming LHP and RHP in quality (Fig. 6, bottom panel). The paper dismisses this by hypothesizing that the classifier memorizes ground-truth motions, but no experiment tests this hypothesis; the alternative that the classifier correctly detects lower realism of the more diverse motions is equally consistent with the data. The human evaluation uses only 8 raters and 50 motions per method, with no confidence intervals or inter-rater agreement statistics, so it is not sufficiently strong to overturn the automatic metric. Therefore the 'high-quality' component of the central claim is not yet established.
  3. [Sections 3.1 and 3.2] The method assumes that any randomly chosen subset of hidden-state entries is informative enough for the decoder; if the information in the RNN hidden state is concentrated in a small number of dimensions, as is often the case in recurrent networks, the random masking will discard that information and quality will suffer. The curriculum schedule mitigates training instability but does not provide any structural guarantee that information is spread uniformly across hidden dimensions. The paper does not analyze this failure mode, so the robustness of the approach across actions and hyperparameters is unclear.
  4. [Section 4.1, Fig. 9] The diversity-quality trade-off plot in Fig. 9 shows that diversity increases with α but quality decreases. For the default α=0.5 used in the main comparisons, the paper does not report the exact quality and diversity values or error bars, making it difficult to judge the operating point against the baselines in Fig. 6. The authors should provide this information to support the claim that α=0.5 yields both high diversity and high quality.
minor comments (7)
  1. [Section 4.1, Fig. 6] The quality subplot does not show error bars, unlike the diversity subplot; please add them or clarify why they are omitted.
  2. [Section 4.1, Fig. 7] It is unclear how the human ratings on a 1-5 scale, scaled to 0-50, are made 'comparable with' the classifier accuracy, which is a percentage; please describe the scaling procedure and the meaning of the axes.
  3. [Section 3.2, Fig. 3] The blocks labeled ResBlock1 and ResBlock2 are not described in the text; please provide their architectures or a reference.
  4. [Section 3.2, last paragraph] The claim that α is the only model-related hyper-parameter is inaccurate because λ (the KL annealing weight) and c (the curriculum index count) also affect training; please revise the statement.
  5. [Section 4.2, bottom of Fig. 10] The text refers to 'LPH' but the abbreviation should be 'LHP' (learned hidden state perturbation).
  6. [Section 3.2] The notation h_t^I and h_t^{\bar{I}} is used without an explicit definition of the complement; please add a sentence explaining that \bar{I} = {1,...,L} \setminus I.
  7. [Section 3.3] The diversity metric as the average pairwise distance between all pairs of generated motions is intuitive, but the paper should state whether the distances are computed in pose space, joint position space, or in the hidden representation space; the implementation details in Section 4.1 mention Euclidean distance but not on which representation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the core claim rests on an architectural mechanism and external comparisons, not on fitted inputs or self-citation.

full rationale

The paper's central claim is that Mix-and-Match perturbation prevents the model from ignoring the random noise, yielding diverse yet high-quality predictions. This is not circular: the random-index replacement is a training/inference mechanism, and the claim that it 'forces' the model is an empirical property verified by comparing decoder-input diversity and output pose diversity against external baselines (LHP, RHP, LPP) under matched architecture and losses. The method does not fit a parameter to a subset of data and then predict a closely related quantity; alpha is a fixed hyperparameter (default 0.5) and the reported diversity/quality values are measured, not derived from the metric definition. The proposed quality metric (classifier accuracy inversely proportional to quality) and diversity metric (average pairwise distance) are author-defined, but they are applied uniformly to all methods and are supplemented by human evaluation and standard MAE/S-MSE comparisons against state-of-the-art deterministic and stochastic models. The paper's self-citations ([1], [2], [29]) appear only in the conclusion as potential applications and are not load-bearing. The only notable weakness is the untested post hoc explanation for the classifier-based quality discrepancy ('the binary classifier memorizing the ground-truth motions'), which is a validity concern for the quality claim, not a circularity in the derivation chain. No equation or fitted quantity is equivalent to the paper's conclusions by construction, and no load-bearing step reduces to a self-citation.

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

The central claim rests on several domain assumptions about human motion multimodality, hidden state redundancy, and the validity of the proposed metrics, plus standard VAE machinery. No physically invented entities are introduced.

free parameters (3)
  • alpha (sampling rate) = 0.5 (default)
    This parameter controls the fraction of the hidden state dimensions that are replaced by random noise. It is chosen by hand and defines the trade-off between quality and diversity in Fig. 9.
  • KL annealing weight lambda = 0 to 1, logistic curve
    This parameter weights the KL divergence term in Eq. (2) and is annealed during training. It is standard CVAE practice, but the specific schedule is a hand-selected design choice.
  • Curriculum index swap count c = 0 to ceil(alpha*L)
    This parameter controls how many hidden state indices are gradually randomized from deterministic to fully random during training. The schedule is detailed only in the missing supplementary material.
assumptions (4)
  • domain assumption The RNN hidden state is sufficiently redundant that any random subset of its dimensions carries enough conditioning information.
    The method replaces a random subset of h_t with standard normal noise; if key information were concentrated in few dimensions, quality would collapse. Curriculum training mitigates but does not guarantee this.
  • domain assumption Human motion prediction is inherently multimodal, so a stochastic model and diversity metric are appropriate.
    The dataset provides one ground-truth future per observation, so the model must learn a distribution from single samples; the paper assumes this is the right problem formulation.
  • standard math The reparameterization trick (Kingma and Welling) is applicable for backpropagation through the CVAE.
    Used in Eq. (1) to sample z_p = mu + Sigma^(1/2) epsilon.
  • ad hoc to paper The binary classifier-based quality metric and average pairwise distance diversity metric are valid proxies for motion quality and diversity.
    These are newly introduced metrics; their validity is assumed when drawing conclusions about state-of-the-art performance, and they may be gameable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Variations in Human Motion via Mix-and-Match Perturbation." pith.science (2026). https://pith.science/paper/WUOGGNMV

@misc{pith2026190800733,
  author       = {Pith},
  title        = {Pith review of: Learning Variations in Human Motion via Mix-and-Match Perturbation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WUOGGNMV}},
  note         = {Machine review of arXiv:1908.00733}
}
read the original abstract

Human motion prediction is a stochastic process: Given an observed sequence of poses, multiple future motions are plausible. Existing approaches to modeling this stochasticity typically combine a random noise vector with information about the previous poses. This combination, however, is done in a deterministic manner, which gives the network the flexibility to learn to ignore the random noise. In this paper, we introduce an approach to stochastically combine the root of variations with previous pose information, which forces the model to take the noise into account. We exploit this idea for motion prediction by incorporating it into a recurrent encoder-decoder network with a conditional variational autoencoder block that learns to exploit the perturbations. Our experiments demonstrate that our model yields high-quality pose sequences that are much more diverse than those from state-of-the-art stochastic motion prediction techniques.

Figures

Figures reproduced from arXiv: 1908.00733 by the authors.

Figure 1
Figure 1. Diversity of K RNN decoder inputs, generated with K = 50 different random vectors. We report the mean diversity over N = 50 samples and the corresponding standard deviation. Recently, several attempts have been made at modeling the stochastic nature of human motion [34, 4, 32, 20, 22]. These methods rely on sampling a random vector that is then combined with an encoding of the observed pose se￾quence. In essence, th… view at source ↗
Figure 2
Figure 2. Mix-and-Match perturbation. (Top) Illustration of the Sampling operation (left) and of the Resampling one (right). Given a sampling rate α and a vector length L, the Sampling opera￾tion samples dαLe indices, say I. The complementary, unsam￾pled, indices are denoted by I¯. Then, given two L-dimensional vectors and the corresponding dαLe and b(1 − α)Lc indices, the Resampling operation mixes the two vectors to form a … view at source ↗
Figure 3
Figure 3. Overview of our approach [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Example of curriculum perturbation of the hidden state. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Architecture of the quality binary classifier. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Quality and diversity evaluation. Our approach outper [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Human (H) and classifier-based (C) evaluation of quality [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Qualitative visualization of diverse motions generated by our model and by the baselines. Each block of columns shows the [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Quality and the diversity of the motions generated with [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Mean angle error (MAE) for the Human 3.6M actions commonly used to report long-term prediction results. [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: Effect of K in the MAE of our actions of the Human3.6M dataset. Note, the bold black one is the best of K = 50 motions, and the shaded area indicates the region between best of K = 1 and K = 500. manner. For instance, our model can be used for human action forecasting…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 22 canonical work pages

  1. [1]

    M. S. Aliakbarian, F. Saleh, B. Fernando, M. Salzmann, L. Petersson, and L. Andersson. Deep action-and context- aware sequence learning for activity recognition and antici- pation. arXiv preprint arXiv:1611.05520, 2016. 9

    work page Pith review arXiv 2016

  2. [2]

    M. S. Aliakbarian, F. S. Saleh, M. Salzmann, B. Fernando, L. Petersson, and L. Andersson. Viena 2: A driving anticipa- tion dataset. In Asian Conference on Computer Vision, pages 449–466. Springer, 2018. 9

    work page 2018

  3. [3]

    J. Bao, D. Chen, F. Wen, H. Li, and G. Hua. Cvae-gan: fine- grained image generation through asymmetric training. In Proceedings of the IEEE International Conference on Com- puter Vision, pages 2745–2754, 2017. 1

    2017

  4. [4]

    Barsoum, J

    E. Barsoum, J. Kender, and Z. Liu. Hp-gan: Probabilistic 3d human motion prediction via gan. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion Workshops, pages 1418–1427, 2018. 1, 2, 3, 5, 6, 7

    work page 2018

  5. [5]

    S. R. Bowman, L. Vilnis, O. Vinyals, A. M. Dai, R. Jozefow- icz, and S. Bengio. Generating sentences from a continuous space. arXiv preprint arXiv:1511.06349, 2015. 2, 4

    arXiv 2015

  6. [6]

    B ¨utepage, H

    J. B ¨utepage, H. Kjellstr ¨om, and D. Kragic. Anticipating many futures: Online human motion prediction and genera- tion for human-robot interaction. In2018 IEEE International Conference on Robotics and Automation (ICRA), pages 1–9. IEEE, 2018. 2

    work page 2018

  7. [7]

    K. Cho, B. Van Merri ¨enboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y . Bengio. Learning phrase representations using rnn encoder-decoder for statistical ma- chine translation. arXiv preprint arXiv:1406.1078, 2014. 5

    arXiv 2014

  8. [8]

    Engel, M

    J. Engel, M. Hoffman, and A. Roberts. Latent constraints: Learning to generate conditionally from unconditional gen- 9 erative models. arXiv preprint arXiv:1711.05772, 2017. 1

    arXiv 2017

Show all 35 references
  1. [9]

    Esser, E

    P. Esser, E. Sutter, and B. Ommer. A variational u-net for conditional appearance and shape generation. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 8857–8866, 2018. 1

    2018

  2. [10]

    Fragkiadaki, S

    K. Fragkiadaki, S. Levine, P. Felsen, and J. Malik. Recurrent network models for human dynamics. In Proceedings of the IEEE International Conference on Computer Vision , pages 4346–4354, 2015. 2, 7, 9

    2015

  3. [11]

    Ghosh, J

    P. Ghosh, J. Song, E. Aksan, and O. Hilliges. Learning hu- man motion models for long-term predictions. In 2017 In- ternational Conference on 3D Vision (3DV), pages 458–466. IEEE, 2017. 2, 9

    2017

  4. [12]

    Gui, Y .-X

    L.-Y . Gui, Y .-X. Wang, X. Liang, and J. M. Moura. Ad- versarial geometry-aware human motion prediction. In Pro- ceedings of the European Conference on Computer Vision (ECCV), pages 786–803, 2018. 1, 2, 7, 8, 9

    2018

  5. [13]

    Gui, Y .-X

    L.-Y . Gui, Y .-X. Wang, D. Ramanan, and J. M. Moura. Few- shot human motion prediction via meta-learning. In Pro- ceedings of the European Conference on Computer Vision (ECCV), pages 432–450, 2018. 2, 7

    2018

  6. [14]

    Ionescu, D

    C. Ionescu, D. Papava, V . Olaru, and C. Sminchisescu. Hu- man3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments. IEEE Transactions on Pattern Analysis and Machine Intelligence , 36(7):1325– 1339, jul 2014. 5

    2014

  7. [15]

    A. Jain, A. R. Zamir, S. Savarese, and A. Saxena. Structural- rnn: Deep learning on spatio-temporal graphs. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5308–5317, 2016. 2, 7, 8, 9

    2016

  8. [16]

    D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. 5

    2014 arXiv

  9. [17]

    D. P. Kingma and M. Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 4

    2013 arXiv

  10. [18]

    T. D. Kulkarni, W. F. Whitney, P. Kohli, and J. Tenenbaum. Deep convolutional inverse graphics network. In Advances in neural information processing systems, pages 2539–2547,

  11. [19]

    Kumar, M

    M. Kumar, M. Babaeizadeh, D. Erhan, C. Finn, S. Levine, L. Dinh, and D. Kingma. Videoflow: A flow-based gen- erative model for video. arXiv preprint arXiv:1903.01434 ,

    1903 arXiv

  12. [20]

    J. N. Kundu, M. Gor, and R. V . Babu. Bihmp-gan: Bidi- rectional 3d human motion prediction gan. arXiv preprint arXiv:1812.02591, 2018. 1, 2

    2018 arXiv

  13. [21]

    A. B. L. Larsen, S. K. Sønderby, H. Larochelle, and O. Winther. Autoencoding beyond pixels using a learned similarity metric. arXiv preprint arXiv:1512.09300, 2015. 1

    2015 arXiv

  14. [22]

    Lin and M

    X. Lin and M. R. Amer. Human motion modeling using dv- gans. arXiv preprint arXiv:1804.10652, 2018. 1, 2

    2018 arXiv

  15. [23]

    Martinez, M

    J. Martinez, M. J. Black, and J. Romero. On human mo- tion prediction using recurrent neural networks. In 2017 IEEE Conference on Computer Vision and Pattern Recog- nition (CVPR), pages 4674–4683. IEEE, 2017. 1, 2, 7, 8, 9

    2017

  16. [24]

    Pascanu, T

    R. Pascanu, T. Mikolov, and Y . Bengio. On the difficulty of training recurrent neural networks. In International confer- ence on machine learning, pages 1310–1318, 2013. 5

    2013

  17. [25]

    Paszke, S

    A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. De- Vito, Z. Lin, A. Desmaison, L. Antiga, and A. Lerer. Auto- matic differentiation in pytorch. 2017. 5

    2017

  18. [26]

    Pavllo, C

    D. Pavllo, C. Feichtenhofer, M. Auli, and D. Grangier. Mod- eling human motion with quaternion-based neural networks. arXiv preprint arXiv:1901.07677, 2019. 1, 2, 4, 7

    1901 arXiv

  19. [27]

    Pavllo, D

    D. Pavllo, D. Grangier, and M. Auli. Quaternet: A quaternion-based recurrent model for human motion. arXiv preprint arXiv:1805.06485, 2018. 1, 2, 4, 7

    2018 arXiv

  20. [28]

    Rodriguez, B

    C. Rodriguez, B. Fernando, and H. Li. Action anticipation by predicting future dynamic images. In European Conference on Computer Vision, pages 89–105. Springer, 2018. 9

    2018

  21. [29]

    Sadegh Aliakbarian, F

    M. Sadegh Aliakbarian, F. Sadat Saleh, M. Salzmann, B. Fer- nando, L. Petersson, and L. Andersson. Encouraging lstms to anticipate actions very early. In The IEEE International Conference on Computer Vision (ICCV), Oct 2017. 9

    2017

  22. [30]

    Y . Shi, B. Fernando, and R. Hartley. Action anticipation with rbf kernelized feature mapping rnn. In Proceedings of the European Conference on Computer Vision (ECCV) , pages 301–317, 2018. 9

    2018

  23. [31]

    K. Sohn, H. Lee, and X. Yan. Learning structured output representation using deep conditional generative models. In Advances in neural information processing systems , pages 3483–3491, 2015. 2

    2015

  24. [32]

    Walker, K

    J. Walker, K. Marino, A. Gupta, and M. Hebert. The pose knows: Video forecasting by generating pose futures. In Computer Vision (ICCV), 2017 IEEE International Confer- ence on, pages 3352–3361. IEEE, 2017. 1, 2, 3, 5, 6

    2017

  25. [33]

    R. J. Williams and D. Zipser. A learning algorithm for con- tinually running fully recurrent neural networks. Neural computation, 1(2):270–280, 1989. 5

    1989

  26. [34]

    X. Yan, A. Rastogi, R. Villegas, K. Sunkavalli, E. Shecht- man, S. Hadap, E. Yumer, and H. Lee. Mt-vae: Learning motion transformations to generate multimodal human dy- namics. In European Conference on Computer Vision, pages 276–293. Springer, 2018. 1, 2, 3, 5, 6, 7

    2018

  27. [35]

    X. Yan, J. Yang, K. Sohn, and H. Lee. Attribute2image: Con- ditional image generation from visual attributes. InEuropean Conference on Computer Vision , pages 776–791. Springer,

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.