DLow: Diversifying Latent Flows for Diverse Human Motion Prediction

Yuan, Ye; Kitani, Kris M.

doi:10.1007/978-3-030-58545-7_20

Cited by 169 publications

(213 citation statements)

References 45 publications

Supporting

Mentioning

198

Contrasting

Order By: Relevance

“…These include feed-forward networks [10,41], convolutional autoencoders [24,17], recurrent neural networks [42,15,16,25], often with adversarial components [17,33], or reinforcement learning [67]. Other approaches include probabilistic models, such as conditional restricted Boltzmann machines [61,60], variational autoencoders [2,3,69,73], and normalizing flows [20]. In general, research in this area assumes that body poses observed in the past are complete-with no missing joints.…”

Section: Related Workmentioning

confidence: 99%

Full-Body Motion from a Single Head-Mounted Device: Generating SMPL Poses from Partial Observations

Dittadi¹,

Dziadzio

Cosker

et al. 2021

2021 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

The increased availability and maturity of head-mounted and wearable devices opens up opportunities for remote communication and collaboration. However, the signal streams provided by these devices (e.g., head pose, hand pose, and gaze direction) do not represent a whole person. One of the main open problems is therefore how to leverage these signals to build faithful representations of the user. In this paper, we propose a method based on variational autoencoders to generate articulated poses of a human skeleton based on noisy streams of head and hand pose. Our approach relies on a model of pose likelihood that is novel and theoretically well-grounded. We demonstrate on publicly available datasets that our method is effective even from very impoverished signals and investigate how pose prediction can be made more accurate and realistic.

show abstract

Section: Related Workmentioning

confidence: 99%

Full-Body Motion from a Single Head-Mounted Device: Generating SMPL Poses from Partial Observations

Dittadi¹,

Dziadzio

Cosker

et al. 2021

2021 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

show abstract

“…We measure the diversity of random markers samples from our WholeGrasp-VAE. We follow [58] to employ the Average L2 Pairwise Distance (APD) to evaluate the diversity within random samples, which is given by AP D =…”

Section: Stochastic Whole-body Grasp Pose Synthesismentioning

confidence: 99%

“…For deterministic models, we measures the marker prediction accuracy by computing the Average L2 Distance Error (ADE) between the predicted markers and ground truth. For our generative model, we follow [58] to measure the sample accuracy by computing the minimal error between the ground truth and 10 random samples. (2) Motion smoothness.…”

Section: Stochastic Motion Infillingmentioning

confidence: 99%

SAGA: Stochastic Whole-Body Grasping with Contact

Wu¹,

Wang²,

Zhang³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Model ADE (m) ↓ FDE (m) ↓ LDS-AF [35] 1.66 3.58 DLow-AF [56] 1.78 3.77 Trajectron++ [46] 1.51 -AgentFormer [57] 1.45…”

Section: B5 Improving Rule-based Plannermentioning

confidence: 99%

“…Prediction performance is compared to reported results for recent state-of-the art models AgentFormer [57], Trajec-tron++ [46], DLow-AF [56], and LDS-AF [35]. Results are shown in Tab.…”

Section: C11 Baseline Comparisonmentioning

confidence: 99%

Generating Useful Accident-Prone Driving Scenarios via a Learned Traffic Prior

Rempe¹,

Philion²,

Guibas³

et al. 2021

Preprint

View full text Add to dashboard Cite

Evaluating and improving planning for autonomous vehicles requires scalable generation of long-tail traffic scenarios. To be useful, these scenarios must be realistic and challenging, but not impossible to drive through safely. In this work, we introduce STRIVE, a method to automatically generate challenging scenarios that cause a given planner to produce undesirable behavior, like collisions. To maintain scenario plausibility, the key idea is to leverage a learned model of traffic motion in the form of a graph-based conditional VAE. Scenario generation is formulated as an optimization in the latent space of this traffic model, effected by perturbing an initial real-world scene to produce trajectories that collide with a given planner. A subsequent optimization is used to find a "solution" to the scenario, ensuring it is useful to improve the given planner. Further analysis clusters generated scenarios based on collision type. We attack two planners and show that STRIVE successfully generates realistic, challenging scenarios in both cases. We additionally "close the loop" and use these scenarios to optimize hyperparameters of a rule-based planner.

show abstract

DLow: Diversifying Latent Flows for Diverse Human Motion Prediction

Cited by 169 publications

References 45 publications

Full-Body Motion from a Single Head-Mounted Device: Generating SMPL Poses from Partial Observations

Full-Body Motion from a Single Head-Mounted Device: Generating SMPL Poses from Partial Observations

SAGA: Stochastic Whole-Body Grasping with Contact

Generating Useful Accident-Prone Driving Scenarios via a Learned Traffic Prior

Contact Info

Product

Resources

About