TrajFlow: A Generative Framework for Occupancy Density Estimation Using Normalizing Flows

Mitch Kosieradzki; SeongJin Choi

REVIEW 4 major objections 6 minor 59 references

TrajFlow: A Generative Framework for Occupancy Density Estimation Using Normalizing Flows

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read TrajFlow claims that predicting each future location's marginal density with a normalizing flow, instead of the joint density of the full trajectory, yields state-of-the-art forecast accuracy and enables fully continuous occupancy sampling.

desk verdict TrajFlow's marginal occupancy density model is a real and useful construction, but the paper's SOTA trajectory-forecasting claim is contradicted by its own Table 1 and an evaluation protocol that rewards incoherent samples. read the letter →

arxiv 2501.14266 v2 pith:3NPOOSSY submitted 2025-01-24 cs.LG

Mitch Kosieradzki , Seongjin Choi This is my paper · ORCID

classification cs.LG

keywords trajectoryforecastingoccupancydensityestimationnormalizingflowsneuralordinarydifferentialequationscontrolledmarginalcontinuousflowmotion

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

TrajFlow is a generative framework for motion forecasting built on one design choice: model each future location's probability distribution separately rather than the joint distribution of the entire future path. The paper argues this marginal formulation is more accurate on standard benchmarks, allows a single model to sample locations at arbitrary future times, and directly produces the two most common motion-forecasting outputs—per-agent trajectories and occupancy grids. The authors implement the idea with neural differential equations in both components—a controlled differential equation as the causal encoder and a continuous normalizing flow as the decoder—and report that this fully continuous version outperforms six baseline models on the ETH/UCY pedestrian benchmarks and on the inD vehicle-intersection dataset. The main cost is compute: training and inference are much slower than a discrete implementation. If the claim is right, one causally encoded flow can serve both trajectory sampling and occupancy mapping for autonomous driving and intersection monitoring.

What carries the argument

The central object is the marginal occupancy density $p(u_s^i \mid O_i,F_i)$, implemented as a conditional normalizing flow. The transform $u_s^i = \Psi_{\psi}^{-1}(\zeta,s,z_s)$ maps a base-latent sample through an invertible network conditioned on the causal embedding and the forecast time $s$; training maximizes exact likelihood through the change-of-variables formula, with a cheap log determinant for affine coupling layers and an integrated trace for the continuous normalizing flow via the instantaneous change of variables. The causal encoder is either a GRU or a neural controlled differential equation driven by a natural cubic spline through the observed trajectory. The load-bearing identity is the factorization in Eq. 11, which makes continuous-time sampling and additive-fusion occupancy grids tractable but also removes explicit modeling of correlations between future positions.

What would settle it

Take a held-out set of real trajectories and compare the likelihood a trained marginal model assigns to each true trajectory under the product formula against the likelihood assigned by a joint flow: if the product model systematically underrates coherent trajectories or overrates zig-zag paths that hop between plausible lanes, the independence assumption is the weak point and the reported minADE gains are an artifact of how the evaluation samples trajectories.

Watch

Extended reading notes

Core claim

The paper's central claim is that the conditional marginal density $p(u_s^i \mid O_i,F_i)$—the probability that agent $i$ occupies location $u_s^i$ at future time $s$ given the observed trajectory $O_i$ and derived features $F_i$—can be learned directly and is a better forecasting target than the joint density of the full unobserved trajectory. The framework factorizes the likelihood of an unobserved path as the product of per-step marginal densities, $P(U \mid O)=\prod_{s=1}^{S} p(u_s \mid O)$, and trains a normalizing flow $\Psi_{\psi}$ that maps a latent sample $z_s$ and time $s$, conditioned on a causal embedding $\zeta = \Phi_{\varphi}(O_i,F_i)$, to a location $u_s^i$. The fully continuous configuration—a neural controlled differential equation encoder and a continuous normalizing flow decoder—achieves the best reported minADE on ETH/UCY and the best RMSE and CRPS on inD among the compared baselines, and the marginal formulation outperforms the paper's own joint formulation on the long-horizon inD data while staying comparable on ETH/UCY.

Load-bearing premise

The framework assumes that, once the observed past is known, each future location is independent of every other future location, and that assumption is what makes the per-step product formula tractable.

Editorial extensions

If this is right

A single trained TrajFlow model can output both trajectory samples and occupancy grids, because the same marginal density can be sampled per time step or fused over time with the additive-fusion rule in Eq. 32.
Occupancy grids can be generated at any sampling frequency, not just the training time step, because forecast time is an explicit input to the flow; the paper shows grids sampled at 10 times the training frequency.
The fully continuous CDE-CNF configuration reaches the reported results with far fewer parameters than the compared baselines: about 60,000 on ETH/UCY and 2.7 million on inD.
On the long-horizon inD experiment, the marginal formulation beats the paper's own joint formulation (RMSE 1.14 vs 1.47; CRPS 0.38 vs 0.45), supporting the claim that marginal densities help long-horizon forecasting.
The computational cost is substantial: the fully continuous model is roughly 19 times slower to train and 11 times slower at inference than the fully discrete GRU-DNF on the pedestrian experiment, so it suits batch or memory-limited settings rather than real-time onboard use.

Reading between the lines

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

Editorial inference: because Eq. 11 drops correlations between future positions, independently sampled time steps produce noisy, incoherent paths—the paper's own top-k sampling is a patch for this. A natural extension is to add a lightweight second-stage temporal model (a copula or small autoregressive head) on top of the marginal flow to restore coherence while keeping the occupancy-density benef
Editorial inference: the marginal formulation's advantage is not uniform—on ETH/UCY the joint version actually has slightly better minADE and minFDE, while on inD the marginal version is clearly better. This suggests the benefit grows with forecast horizon and scene complexity, so the strongest test of the paper's idea would be other long-horizon vehicle datasets.
Editorial inference: a calibrated marginal density can be used directly for collision-probability estimates and black-spot identification, as the paper suggests but does not test; a concrete follow-up would compare TrajFlow-derived occupancy maps against existing black-spot methods on road-safety data.
Editorial inference: the ETH/UCY results sit close to what the paper calls the entropy floor of the dataset, so the practical improvement over baselines on pedestrian scenes is small; the cleaner evidence for the marginal design is the comparison against the paper's own joint baseline under the same architecture.

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.

Desk Editor's Note

read the letter

Thanks for the report. I read TrajFlow alongside your notes. I think the stress-test is right: the paper's headline claim about trajectory accuracy doesn't survive its own results. Table 1 has CDE-CNF (Joint) at 0.18/0.37 and CDE-CNF (Marginal) at 0.19/0.38 on ETH/UCY, so the marginal formulation is worse on the trajectory metrics there. The abstract's 'higher accuracy' and the contribution bullet claiming state-of-the-art performance are overstated. And your point about Eq. 11 is the mechanism: sampling each future position independently and then scoring with minADE/minFDE over 20 draws lets the marginal model pick the best location at every time step without committing to a coherent path. That is a real evaluation mismatch.

What is actually new: modeling the per-time marginal spatial density with a normalizing flow conditioned on a neural-CDE embedding, with continuous-time sampling via a CNF. I don't see that exact combination in FloMo, Trajectron++, or SocialVAE. The paper also ships code, runs clean ablations across four configurations, and is honest in the Discussion about the joint/marginal trade-off and the diversity limitation. The inD results (Table 2) show the marginal CDE-CNF beating the joint on RMSE/CRPS, but those are per-point marginal metrics, so they support the occupancy-density claim, not the trajectory-forecasting claim. Citation pattern is clean: the only self-citation is the second author's tutorial [34], used as background. No circularity in the empirical comparisons.

Soft spots in order of importance. First, the evaluation protocol: inD uses a single recording session split with no error bars, and only six baselines, so 'state-of-the-art' is not established. Second, the conditional-independence factorization is a strong assumption for trajectory generation; the paper acknowledges reduced diversity, but the evaluation still treats the samples as trajectories. Third, the runtime costs are severe (CDE-CNF is 19x slower to train than GRU-DNF), which undercuts the practical contribution unless the occupancy-grid use case is the target. None of these sink the core idea; they mean the claims need to be revised.

Who this is for: researchers working on occupancy grids and continuous-time trajectory density models, especially in motion forecasting. The paper deserves a serious referee: the construction is coherent, the code is public, and the marginal-vs-joint question is worth airing. My recommendation: send it to review, but the authors should be asked to fix the abstract, re-run or re-report the trajectory comparison with a standard inD protocol, and add a discussion of the sampling-diversity/evaluation mismatch. If they can't soften the claims, the paper is still a decent workshop-level contribution, but as written the central accuracy claim is not supportable.

Referee Report

4 major / 6 minor

Summary. The paper proposes TrajFlow, a generative framework for occupancy density estimation. The model uses a causal encoder (GRU or neural CDE) to embed an observed trajectory and a normalizing flow (affine couplings or neural ODE) to model the conditional marginal density of each future spatial location p(u_s | O, F). The authors claim that modeling marginal rather than joint trajectory densities yields higher forecasting accuracy, enables continuous-time sampling, and supports both trajectory sampling and occupancy grids. They evaluate discrete and continuous implementations on ETH/UCY and inD, reporting ablations and comparisons against six baselines.

Significance. If the claims were fully supported, the paper would provide a useful contribution to motion forecasting by unifying trajectory and occupancy representations in a single flow-based model. The code is publicly available, and the systematic ablation of discrete versus continuous encoders and flows is informative. The core methodological idea of marginal occupancy density estimation is worthwhile. However, the paper's central claim that the marginal formulation produces higher trajectory-forecasting accuracy is not supported by its own evidence, and the evaluation protocol for trajectory metrics is mismatched with the marginal sampling procedure. The contribution is better positioned as occupancy-density estimation than as superior trajectory prediction.

major comments (4)

[Abstract; Table 1; Discussion] The abstract and contribution list claim that the marginal formulation produces higher accuracy on trajectory forecasting benchmarks, but Table 1 shows CDE-CNF (Marginal) at 0.19/0.38 versus CDE-CNF (Joint) at 0.18/0.37 on ETH/UCY, and the Discussion itself states that 'the marginal and joint formulations produced comparable results on the ETH/UCY dataset.' This internal contradiction must be reconciled; either the claims should be revised or additional evidence should be provided to support the stated advantage.
[Eq. (11) and Evaluation Metrics] Because the model generates trajectories by independently sampling each future position from the marginal density (Eq. 11), the minADE/minFDE metrics, which select the best of 20 independently drawn samples, can reward temporally incoherent trajectories: the best sample can pick the most probable location at each time step without forming a physically plausible path. The inD results (Table 2) use RMSE/CRPS, which are per-point marginal metrics and do not assess trajectory coherence. To support the trajectory-forecasting claim, the authors should evaluate with trajectory-level metrics that respect temporal coupling, for example by sampling with a coherence-enforcing strategy (such as their top-k sampling) or by measuring the error of full sampled trajectories without per-timestep recombination.
[Vehicle Experiment] The inD evaluation uses a single random 75/25 split of one recording session, which is nonstandard and does not permit any assessment of variance; this makes the reported state-of-the-art claim fragile. The authors should adopt the standard inD benchmark splits or report results across multiple seeds/splits with confidence intervals, and should clarify how the reported baseline numbers were obtained under the same protocol.
[Discussion and Figure 6] The Discussion concedes that the marginal formulation 'inherently reduces diversity among the sampled motion trajectories and introduces noise, as the relationships between predicted locations are no longer explicitly modeled,' and Figure 6 shows that a top-k sampling strategy is needed to reduce the resulting noise. This admission directly undermines the abstract's claim that the marginal formulation produces higher trajectory-forecasting accuracy, and it should be reflected in the paper's framing: the contribution is better positioned as occupancy-density estimation, not as superior trajectory prediction.

minor comments (6)

[Experiments (pedestrian dataset description)] The text calls the dataset 'UTY' but it should be 'UCY'; please correct this typo.
[Baseline Comparison (inD)] In the paragraph discussing inD baseline comparison, 'MSE' should be 'RMSE' to match the metric defined in the Evaluation Metrics subsection.
[Methodology / implementation details] The paper does not provide sufficient architecture details for reproducibility, such as the number of affine coupling layers, hidden sizes and activation functions for the MLPs in the coupling layers and the CDE vector field, the FiLM layer dimensions, and training hyperparameters like batch size; please add these details.
[Discussion, Eq. (32)] Equation (32) is not typeset correctly; the max over u_i,s is missing its intended subscript and the expression should be defined more carefully so the additive fusion procedure is unambiguous.
[Methodology, Eq. (27)] The text calls Eq. (27) the 'concat and squash layer', but the equation describes FiLM conditioning; the terminology is confusing and should be aligned with the FiLM citation.
[Contributions, item 4] The claim of state-of-the-art performance should be qualified: in Table 1, CDE-CNF does not beat PPT on minFDE (0.38 vs. 0.31), so the state-of-the-art statement is too strong as written.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the central claim rests on out-of-sample comparisons, and the architecture is built on external prior work; the only self-citation is a non-load-bearing background tutorial.

full rationale

TrajFlow's derivation chain is self-contained rather than circular. The key likelihood in Eq. 11 defines the joint trajectory likelihood as a product of per-step marginals under conditional independence, and Eq. 12 defines the generative map from a latent sample through the inverse flow. This is a modeling assumption, not a result derived from the outputs being predicted. The accuracy claims are supported by leave-one-out evaluation on ETH/UCY and a held-out split of inD against published baselines (Tables 1 and 2), so no fitted parameter is relabeled as a prediction and no benchmark quantity reduces by construction to a training target. The only self-citation is reference [34] (Choi et al., a tutorial on deep generative models in transportation), used in the background sentence about generative modeling; it is not load-bearing for any technical claim. The paper's own admission of reduced trajectory diversity under the marginal assumption is a limitation, not a circular step. The evaluation-metric concern raised elsewhere (minADE on independently sampled marginals rewarding incoherent trajectories) is a validity/correctness issue, not a circularity issue, and the internal joint-vs-marginal comparison in Table 1 is reported honestly. Accordingly, no circular step meets the evidence bar set by the review criteria.

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

The main results rest on standard flow and ODE mathematics (change of variables, instantaneous change of variables, adjoint method) and on a domain-specific factorization assumption (Eq. 11). The only nonstandard load-bearing assumption is conditional independence across future timesteps. No new physical entities are introduced; 'additive fusion' is an inference-time aggregation procedure rather than an invented entity. Several hyperparameters (augmentation range, normalization bounds, evaluation sample counts) are chosen by hand and affect the reported numbers.

free parameters (4)

Random scaling augmentation range on ETH/UCY = [0.3, 1.7]
Chosen by hand to augment trajectories; affects training distribution and reported results. Not derived from data or theory.
inD min-max normalization bounds = computed from training split
x_min and x_max are fitted to the selected inD training recording and applied to test; a data-dependent preprocessing choice.
Evaluation sampling budget = 20 samples for ETH/UCY; 1,000 for inD
The number of Monte Carlo samples changes minADE/minFDE and RMSE/CRPS values; chosen to match prior work for ETH/UCY and ad hoc for inD.
Architecture and solver hyperparameters = not fully specified
Layer counts, widths, and ODE solver tolerances (1e-5) are choices made during development; not enough detail is given to treat them as fixed or derived.

assumptions (4)

domain assumption Conditional independence of future unobserved positions (Eq. 11): P(U|O) = prod_s p(u_s|O).
Makes the joint likelihood a product of marginal densities. If false, trajectory samples are incoherent and the reported edge of the marginal formulation may be an artifact of evaluation.
standard math Instantaneous change of variables for CNF (Eq. 8) and standard change of variables for discrete flows (Eq. 2).
Background from Chen et al. 2018 and Grathwohl et al. 2018; used without proof.
domain assumption Natural cubic spline interpolation of observed trajectory and features provides an adequate continuous control path for the CDE.
The choice of interpolation scheme influences the encoder dynamics; no evidence is given that it is optimal or unbiased.
domain assumption Leave-one-out ETH/UCY and single-recording-session inD splits produce representative evaluation.
The inD split is nonstandard and may not generalize to other recordings or intersections.

how reviews work

0 comments

Cite this review

Pith. "Pith review of TrajFlow: A Generative Framework for Occupancy Density Estimation Using Normalizing Flows." pith.science (2026). https://pith.science/paper/3NPOOSSY

@misc{pith2026250114266,
  author       = {Pith},
  title        = {Pith review of: TrajFlow: A Generative Framework for Occupancy Density Estimation Using Normalizing Flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3NPOOSSY}},
  note         = {Machine review of arXiv:2501.14266}
}

read the original abstract

For intelligent transportation systems and autonomous vehicles to operate safely and efficiently, they must reliably predict the future motion and trajectory of surrounding agents within complex traffic environments. At the same time, the motion of these agents is inherently uncertain, making accurate prediction difficult. In this paper, we propose \textbf{TrajFlow}, a generative framework for estimating the occupancy density of dynamic agents. Our framework utilizes a causal encoder to extract semantically meaningful embeddings of the observed trajectory, as well as a normalizing flow to decode these embeddings and determine the most likely future location of an agent at some time point in the future. Our formulation differs from existing approaches because we model the marginal distribution of spatial locations instead of the joint distribution of unobserved trajectories. The advantages of a marginal formulation are numerous. First, we demonstrate that the marginal formulation produces higher accuracy on challenging trajectory forecasting benchmarks. Second, the marginal formulation allows for fully continuous sampling of future locations. Finally, marginal densities are better suited for downstream tasks as they allow for the computation of per-agent motion trajectories and occupancy grids, the two most commonly used representations for motion forecasting. We present a novel architecture based entirely on neural differential equations as an implementation of this framework and provide ablations to demonstrate the advantages of a continuous implementation over a more traditional discrete neural network based approach. The code is available at https://github.com/UMN-Choi-Lab/TrajFlow.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 49 canonical work pages

[34]

Choi, S., Z. Jin, S. Ham, J. Kim, and L. Sun, A Gentle Introduction and Tutorial on Deep Generative Models in Transportation Research. arXiv preprint arXiv:2410.07066, 2024

arXiv 2024
[1]

Tyagi, A. K. and N. Sreenath, Introduction to intelligent transportation system. In Intelli- gent transportation systems: theory and practice, Springer, 2022, pp. 1–22

work page 2022
[2]

Alahi, A., K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese, Social LSTM: Human Trajectory Prediction in Crowded Spaces. In Proceedings of the Confer- ence on Computer Vision and Pattern Recognition (CVPR) , IEEE, Las Vegas, NV , USA, 2016, pp. 961–971

work page 2016
[3]

Ouyang, P

Zhang, P., W. Ouyang, P. Zhang, J. Xue, and N. Zheng, SR-LSTM: State Refinement for LSTM Towards Pedestrian Trajectory Prediction. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Long Beach, CA, USA, 2019, pp. 1203–1212

work page 2019
[4]

Muelling, and J

Vemula, A., K. Muelling, and J. Oh, Social Attention: Modeling Attention in Human Crowds. In Proceedings of the International Conference on Robotics and Automation (ICRA), IEEE, Brisbane, Australia, 2018, pp. 5594–5601

work page 2018
[5]

Pajouheshgar, E. and C. H. Lampert, Back to Square One: Probabilistic Trajectory Fore- casting Without Bells and Whistles. In Proceedings of the Conference on Neural Informa- tion Processing Systems Workshops (NeurIPS Workshops) , NeurIPS, Montreal, Canada, 2018

work page 2018
[6]

Ji, S., W. Xu, M. Yang, and K. Yu, 3D Convolutional Neural Networks for Human Action Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), V ol. 35, No. 1, 2013, pp. 221–231

work page 2013
[7]

Ivanovic, P

Salzmann, T., B. Ivanovic, P. Chakravarty, and M. Pavone, Trajectron++: Dynamically- feasible trajectory forecasting with heterogeneous data. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVIII 16, Springer, 2020, pp. 683–700

work page 2020

Show all 59 references

[8]

Schöller, C. and A. Knoll, Flomo: Tractable motion prediction with normalizing flows. In 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2021, pp. 7977–7984

2021
[9]

Tabak, E. G. and C. V . Turner, A family of nonparametric density estimation algorithms. Communications on Pure and Applied Mathematics, V ol. 66, No. 2, 2013, pp. 145–164. Kosieradzki and Choi 21

2013
[10]

Höffken, U

Wiest, J., M. Höffken, U. Kreßel, and K. Dietmayer, Probabilistic trajectory prediction with Gaussian mixture models. In 2012 IEEE Intelligent Vehicles Symposium , 2012, pp. 141–146

2012
[11]

arXiv preprint arXiv:1312.6114, 2013

Kingma, D., Auto-Encoding Variational Bayes. arXiv preprint arXiv:1312.6114, 2013

2013 arXiv
[12]

Pouget-Abadie, M

Goodfellow, I., J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, Generative adversarial networks. Communications of the ACM, V ol. 63, No. 11, 2020, pp. 139–144

2020
[13]

Thrun, S. and A. Bücken, Integrating Grid-Based and Topological Maps for Mobile Robot Navigation. In Proceedings of the National Conference on Artificial Intelligence, 1996, pp. 944–951

1996
[14]

Ghadi, M. and Á. Török, A comparative analysis of black spot identification methods and road accident segmentation methods. Accident Analysis & Prevention, V ol. 128, 2019, pp. 1–7

2019
[15]

Ghadi, M. and Á. Török, Comparison different black spot identification methods. Trans- portation research procedia, V ol. 27, 2017, pp. 1105–1112

2017
[16]

Halekoh, W

Debrabant, B., U. Halekoh, W. H. Bonat, D. L. Hansen, J. Hjelmborg, and J. Lauritsen, Identifying traffic accident black spots with Poisson-Tweedie models. Accident Analysis & Prevention, V ol. 111, 2018, pp. 147–154

2018
[17]

Wiyono, et al., Three strategies reducing accident rates at black spots and black sites road in Riau Province, Indonesia

Sandhyavitri, A., S. Wiyono, et al., Three strategies reducing accident rates at black spots and black sites road in Riau Province, Indonesia. Transportation research procedia, V ol. 25, 2017, pp. 2153–2166

2017
[18]

Chen, R. T., Y . Rubanova, J. Bettencourt, and D. K. Duvenaud, Neural ordinary differential equations. Advances in neural information processing systems, V ol. 31, 2018

2018
[19]

Morrill, J

Kidger, P., J. Morrill, J. Foster, and T. Lyons, Neural controlled differential equations for irregular time series. Advances in Neural Information Processing Systems, V ol. 33, 2020, pp. 6696–6707

2020
[20]

LeCun, Y . and Y . Bengio, Convolutional networks for images, speech, and time series. In The Handbook of Brain Theory and Neural Networks , MIT Press, V ol. 3361, 1995, p. 1995

1995
[21]

Sutskever, and G

Krizhevsky, A., I. Sutskever, and G. E. Hinton, ImageNet Classification with Deep Con- volutional Neural Networks. In Advances in Neural Information Processing Systems 25 , NeurIPS, 2012, conference Paper

2012
[22]

Zhang, S

He, K., X. Zhang, S. Ren, and J. Sun, Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

2016
[23]

Huang, G., Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, Densely connected convo- lutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 4700–4708

2017
[24]

Donahue, T

Girshick, R., J. Donahue, T. Darrell, and J. Malik, Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 580–587

2014
[25]

Divvala, R

Redmon, J., S. Divvala, R. Girshick, and A. Farhadi, You only look once: Unified, real- time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 779–788. Kosieradzki and Choi 22

2016
[26]

Shelhamer, and T

Long, J., E. Shelhamer, and T. Darrell, Fully convolutional networks for semantic segmen- tation. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3431–3440

2015
[27]

Fischer, and T

Ronneberger, O., P. Fischer, and T. Brox, U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention– MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, pro- ceedings, part III 18,...

2015
[28]

arXiv preprint arXiv:1406.1078, 2014

Cho, K., Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation. arXiv preprint arXiv:1406.1078, 2014

2014 arXiv
[29]

Vinyals, and Q

Sutskever, I., O. Vinyals, and Q. V . Le, Sequence to sequence learning with neural net- works. Advances in neural information processing systems, V ol. 27, 2014

2014
[30]

arXiv preprint arXiv:1308.0850, 2013

Graves, A., Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850, 2013

2013 arXiv
[31]

Flunkert, J

Salinas, D., V . Flunkert, J. Gasthaus, and T. Januschowski, DeepAR: Probabilistic forecast- ing with autoregressive recurrent networks. International journal of forecasting, V ol. 36, No. 3, 2020, pp. 1181–1191

2020
[32]

Johnson, L

Gupta, A., J. Johnson, L. Fei-Fei, S. Savarese, and A. Alahi, Social GAN: Socially Accept- able Trajectories with Generative Adversarial Networks. InProceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2018, pp. 2255–2264

2018
[33]

Hayet, and I

Xu, P., J.-B. Hayet, and I. Karamouzas, SocialV AE: Human Trajectory Prediction using Timewise Latents. In European Conference on Computer Vision, Springer, 2022, pp. 511– 528

2022
[35]

Tata McGraw-Hill Education, New Delhi, 2006

Rudin, W., Real and Complex Analysis. Tata McGraw-Hill Education, New Delhi, 2006

2006
[36]

I., Measure Theory

Bogachev, V . I., Measure Theory. Springer Berlin Heidelberg, Berlin, 2007

2007
[37]

Zhong, Q

Lu, Y ., A. Zhong, Q. Li, and B. Dong, Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations. In International Conference on Machine Learning, PMLR, 2018, pp. 3276–3285

2018
[38]

Haber, E. and L. Ruthotto, Stable architectures for deep neural networks. Inverse problems, V ol. 34, No. 1, 2017, p. 014004

2017
[39]

Ruthotto, L. and E. Haber, Deep neural networks motivated by partial differential equa- tions. Journal of Mathematical Imaging and Vision, V ol. 62, No. 3, 2020, pp. 352–364

2020
[40]

S., Mathematical Theory of Optimal Processes

Pontryagin, L. S., Mathematical Theory of Optimal Processes. Routledge, 1962

1962
[41]

Grathwohl, W., R. T. Chen, J. Bettencourt, I. Sutskever, and D. Duvenaud, Ffjord: Free-form continuous dynamics for scalable reversible generative models. arXiv preprint arXiv:1810.01367, 2018

2018 arXiv
[42]

F., A Stochastic Estimator of the Trace of the Influence Matrix for Lapla- cian Smoothing Splines.Journal of Computational and Graphical Statistics, V ol

Hutchinson, M. F., A Stochastic Estimator of the Trace of the Influence Matrix for Lapla- cian Smoothing Splines.Journal of Computational and Graphical Statistics, V ol. 18, 1989, pp. 1059–1076, doi:10.1080/10618600.1989.10474754

1989
[43]

Sohl-Dickstein, and S

Dinh, L., J. Sohl-Dickstein, and S. Bengio, Density estimation using real nvp. arXiv preprint arXiv:1605.08803, 2016

2016 arXiv
[44]

Krueger, and Y

Dinh, L., D. Krueger, and Y . Bengio, Nice: Non-linear independent components estima- tion. arXiv preprint arXiv:1410.8516, 2014. Kosieradzki and Choi 23

2014 arXiv
[45]

Ioffe, S. and C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167, 2015

2015 arXiv
[46]

Bartels, R. H., J. C. Beatty, and B. A. Barsky, Hermite and Cubic Spline Interpolation. In An Introduction to Splines for Use in Computer Graphics and Geometric Modelling (W. Carlson and G. Goos, eds.), Morgan Kaufmann, San Francisco, CA, 1998, pp. 9–17

1998
[47]

Strub, H

Perez, E., F. Strub, H. De Vries, V . Dumoulin, and A. Courville, Film: Visual reasoning with a general conditioning layer. In Proceedings of the AAAI conference on artificial intelligence, 2018, V ol. 32

2018
[48]

Pellegrini, S., A. Ess, K. Schindler, and L. V . Gool, You’ll never walk alone: Modeling social behavior for multi-target tracking. In Proceedings of the International Conference on Computer Vision (ICCV), IEEE, 2009, pp. 261–268

2009
[49]

Chrysanthou, and D

Lerner, A., Y . Chrysanthou, and D. Lischinski, Crowds by Example. In Computer Graphics Forum, Wiley Online Library, 2007, V ol. 26, pp. 655–664

2007
[50]

Krajewski, T

Bock, J., R. Krajewski, T. Moers, S. Runde, L. Vater, and L. Eckstein, The ind dataset: A drone dataset of naturalistic road user trajectories at german intersections. In 2020 IEEE Intelligent Vehicles Symposium (IV), IEEE, 2020, pp. 1929–1934

2020
[51]

P., Adam: A method for stochastic optimization

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

2014 arXiv
[52]

Dormand, J. and P. Prince, Runge-Kutta methods for all ordinary differential equations. Journal of Computational and Applied Mathematics, V ol. 6, No. 1, 1980, pp. 19–26

1980
[53]

Shi, L., L. Wang, S. Zhou, and G. Hua, Trajectory Unified Transformer for Pedestrian Trajectory Prediction. In Proceedings of the IEEE/CVF International Conference on Com- puter Vision (ICCV), 2023, pp. 9675–9684

2023
[54]

Liang, J

Lin, X., T. Liang, J. Lai, and J.-F. Hu, Progressive pretext task learning for human trajec- tory prediction. In European Conference on Computer Vision , Springer, 2024, pp. 197– 214

2024
[55]

Graves, A. and A. Graves, Long short-term memory. Supervised sequence labelling with recurrent neural networks, 2012, pp. 37–45

2012
[56]

Bekasov, I

Durkan, C., A. Bekasov, I. Murray, and G. Papamakarios, Neural spline flows. Advances in neural information processing systems, V ol. 32, 2019

2019
[57]

Lambert, P

Chang, M.-F., J. Lambert, P. Sangkloy, J. Singh, S. Bak, A. Hartnett, D. Wang, P. Carr, S. Lucey, D. Ramanan, et al., Argoverse: 3d tracking and forecasting with rich maps. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 8748–8757

2019
[58]

Matheson, J. E. and R. L. Winkler, Scoring Rules for Continuous Probability Distributions. Management Science, V ol. 22, No. 10, 1976, pp. 1087–1096

1976
[59]

Mahjourian, R., J. Kim, Y . Chai, M. Tan, B. Sapp, and D. Anguelov, Occupancy flow fields for motion forecasting in autonomous driving. IEEE Robotics and Automation Letters , V ol. 7, No. 2, 2022, pp. 5639–5646

2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [34]

Choi, S., Z. Jin, S. Ham, J. Kim, and L. Sun, A Gentle Introduction and Tutorial on Deep Generative Models in Transportation Research. arXiv preprint arXiv:2410.07066, 2024

arXiv 2024

[2] [1]

Tyagi, A. K. and N. Sreenath, Introduction to intelligent transportation system. In Intelli- gent transportation systems: theory and practice, Springer, 2022, pp. 1–22

work page 2022

[3] [2]

Alahi, A., K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese, Social LSTM: Human Trajectory Prediction in Crowded Spaces. In Proceedings of the Confer- ence on Computer Vision and Pattern Recognition (CVPR) , IEEE, Las Vegas, NV , USA, 2016, pp. 961–971

work page 2016

[4] [3]

Ouyang, P

Zhang, P., W. Ouyang, P. Zhang, J. Xue, and N. Zheng, SR-LSTM: State Refinement for LSTM Towards Pedestrian Trajectory Prediction. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Long Beach, CA, USA, 2019, pp. 1203–1212

work page 2019

[5] [4]

Muelling, and J

Vemula, A., K. Muelling, and J. Oh, Social Attention: Modeling Attention in Human Crowds. In Proceedings of the International Conference on Robotics and Automation (ICRA), IEEE, Brisbane, Australia, 2018, pp. 5594–5601

work page 2018

[6] [5]

Pajouheshgar, E. and C. H. Lampert, Back to Square One: Probabilistic Trajectory Fore- casting Without Bells and Whistles. In Proceedings of the Conference on Neural Informa- tion Processing Systems Workshops (NeurIPS Workshops) , NeurIPS, Montreal, Canada, 2018

work page 2018

[7] [6]

Ji, S., W. Xu, M. Yang, and K. Yu, 3D Convolutional Neural Networks for Human Action Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), V ol. 35, No. 1, 2013, pp. 221–231

work page 2013

[8] [7]

Ivanovic, P

Salzmann, T., B. Ivanovic, P. Chakravarty, and M. Pavone, Trajectron++: Dynamically- feasible trajectory forecasting with heterogeneous data. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVIII 16, Springer, 2020, pp. 683–700

work page 2020