MG-GAN: A Multi-Generator Model Preventing Out-of-Distribution Samples in Pedestrian Trajectory Prediction

@article{Dendorfer2021MGGANAM,
  title={MG-GAN: A Multi-Generator Model Preventing Out-of-Distribution Samples in Pedestrian Trajectory Prediction},
  author={Patrick Dendorfer and Sven Elflein and Laura Leal-Taix{\'e}},
  journal={2021 IEEE/CVF International Conference on Computer Vision (ICCV)},
  year={2021},
  pages={13138-13147}
}
Pedestrian trajectory prediction is challenging due to its uncertain and multimodal nature. While generative adversarial networks can learn a distribution over future trajectories, they tend to predict out-of-distribution samples when the distribution of future trajectories is a mixture of multiple, possibly disconnected modes. To address this issue, we propose a multi-generator model for pedestrian trajectory prediction. Each generator specializes in learning a distribution over trajectories… 

PedFormer: Pedestrian Behavior Prediction via Cross-Modal Attention Modulation and Gated Multitask Learning

This work proposes a novel framework that relies on different data modalities to predict future trajectories and crossing actions of pedestrians from an ego-centric perspective and demonstrates that this model improves state-of-the-art in trajectory and action prediction by up to 22% and 13% respectively on various metrics.

LatentFormer: Multi-Agent Transformer-Based Interaction Modeling and Trajectory Prediction

LatentFormer, a transformerbased model for predicting future vehicle trajectories, is proposed that leverages a novel technique for modeling interactions among dynamic objects in the scene and achieves state-of-the-art performance and improves upon trajectory metrics by up to 40%.

Learning Pedestrian Group Representations for Multi-modal Trajectory Prediction

GP-Graph is presented, which has collective group representations for effective pedestrian trajectory prediction in crowded environments, and is compatible with all types of existing approaches, to model both individual-wise and group-wise relations as graph representations.

Non-Probability Sampling Network for Stochastic Human Trajectory Prediction

The Non-Probability Sampling Network (NPSN), a very small network that generates purposive sample sequences using the past paths of pedestrians and their social interactions, is proposed, which can significantly improve both the prediction accuracy and reliability of the public pedestrian trajectory prediction benchmark.

Adaptive Trajectory Prediction via Transferable GNN

A novel Transferable Graph Neural Network frame-work is proposed, which jointly conducts trajectory prediction as well as domain alignment in a unified framework and an attention-based adaptive knowledge learning module is further proposed to explore fine-grained individual-level feature representations for knowledge transfer.

Goal-driven Self-Attentive Recurrent Networks for Trajectory Prediction

This work proposes a lightweight attention-based recurrent backbone that acts solely on past observed positions, based on a U-Net architecture, and demonstrates that its prediction accuracy can be improved considerably when combined with a scene-aware goal-estimation module.

Stochastic Trajectory Prediction via Motion Indeterminacy Diffusion

This paper presents a new framework to formulate the trajectory prediction task as a reverse process of motion indeterminacy diffusion (MID), in which the history behavior information and the social interactions are encoded as a state embedding and a Transformer-based diffusion model is devised to capture the temporal dependencies of trajectories.

Dynamic-Group-Aware Networks for Multi-Agent Trajectory Prediction with Relational Reasoning

DynGroupNet is proposed, a dynamic-group-aware network, which can model time-varying interactions in highly dynamic scenes, and capture both pair-wise and group-wise interactions and reason both interaction strength and category without direct supervision during trajectory prediction.

Collaborative Uncertainty Benefits Multi-Agent Multi-Modal Trajectory Forecasting

A novel concept, collaborative uncertainty (CU), which models the uncertainty resulting from interaction modules is proposed, and a general CU-aware regression framework with an original permutation-equivariant uncertainty estimator is built to do both tasks of regression and uncertainty estimation.

Human Action Recognition and Prediction: A Survey

The complete state-of-the-art techniques in the action recognition and prediction are surveyed, including existing models, popular algorithms, technical difficulties, popular action databases, evaluation protocols, and promising future directions are provided.

References

SHOWING 1-10 OF 45 REFERENCES

Social Ways: Learning Multi-Modal Distributions of Pedestrian Trajectories With GANs

This paper proposes a novel approach for predicting the motion of pedestrians interacting with others that uses a Generative Adversarial Network to sample plausible predictions for any agent in the scene and shows that the proposed method leads to generate more diverse samples and to preserve the modes of the predictive distribution.

Conditional Generative Neural System for Probabilistic Trajectory Prediction

This paper proposes a conditional generative neural system (CGNS) for probabilistic trajectory prediction to approximate the data distribution, with which realistic, feasible and diverse future trajectory hypotheses can be sampled.

Social-BiGAT: Multimodal Trajectory Forecasting using Bicycle-GAN and Graph Attention Networks

A graph-based generative adversarial network that generates realistic, multimodal trajectory predictions by better modelling the social interactions of pedestrians in a scene and achieves state-of-the-art performance comparing it to several baselines on existing trajectory forecasting benchmarks.

Social LSTM: Human Trajectory Prediction in Crowded Spaces

This work proposes an LSTM model which can learn general human movement and predict their future trajectories and outperforms state-of-the-art methods on some of these datasets.

Multi-agent Diverse Generative Adversarial Networks

MAD-GAN, an intuitive generalization to the Generative Adversarial Networks and its conditional variants to address the well known problem of mode collapse is proposed and its efficacy on the unsupervised feature representation task is shown.

Trajectory Forecasts in Unknown Environments Conditioned on Grid-Based Plans

An attention based trajectory generator that generates continuous valued future trajectories conditioned on state sequences sampled from the MaxEnt policy is proposed that generates trajectories that are diverse, representing the multi-modal predictive distribution, and precise, conforming to the underlying scene structure over long prediction horizons.

Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks

A recurrent sequence-to-sequence model observes motion histories and predicts future behavior, using a novel pooling mechanism to aggregate information across people, and outperforms prior work in terms of accuracy, variety, collision avoidance, and computational complexity.

MGAN: Training Generative Adversarial Nets with Multiple Generators

A new approach to train the Generative Adversarial Nets with a mixture of generators to overcome the mode collapsing problem, and develops theoretical analysis to prove that, at the equilibrium, the Jensen-Shannon divergence (JSD) between the mixture of generator’ distributions and the empirical data distribution is minimal, whilst the JSD among generators' distributions is maximal, hence effectively avoiding the mode collapse problem.

Disconnected Manifold Learning for Generative Adversarial Networks

This work shows that single generator GANs are unable to correctly model a distribution supported on a disconnected manifold, and investigates how sample quality, mode dropping and local convergence are affected by this.

Generative Adversarial Nets

We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a