TrajNet++: Large-scale Trajectory Forecasting Benchmark

Published in 2nd Workshop on Long-Term Human Motion Prediction, ICRA, 2020

Current forecasting methods have been evaluated on different subsets of the available data without proper indexing of trajectories making it difficult to objectively compare the techniques. Our benchmark provides defined categorization of trajectories as well as a unified extensive evaluation system to test the gathered methods for a fair comparison.

Recommended citation: Kothari, Parth, Sven Kreiss and Alexandre Alahi. “Human Trajectory Forecasting in Crowds: A Deep Learning Perspective.” (2020). arxiv preprint arXiv:2007.03639.

Collaborative Sampling in Generative Adversarial Networks

Published in Association for Advancement of Artificial Intelligence (AAAI), 2020

Developed a collaborative sampling scheme between the generator and the discriminator for improved data generation during sampling. Proposed a practical discriminator shaping method for effective sample refinement. Experiments on synthetic and image datasets demonstrate the efficacy of our method to improve generated samples both quantitatively and qualitatively, offering a new degree of freedom in GAN sampling.

Recommended citation: Liu, Yuejiang, Parth Kothari and Alexandre Alahi. “Collaborative Sampling in Generative Adversarial Networks.” Association for Advancement of Artificial Intelligence (AAAI) 2020.

Adversarial Loss in Human Trajectory Prediction

Published in European Association for Research in Transportation (hEART), 2019

Highlighted an unexpected pitfall in the state-of-the-art architecture for multimodal human prediction via controlled experiments. Proposed a modification to the architecture leveraging the progress in the GAN community. Demonstrate the efficacy of the proposed modification on real world datasets, indicating room for improvement on state-of-the-art.

Recommended citation: Parth Kothari, Alexandre Alahi (2019). "Adversarial Loss in Human Trajectory Prediction." European Association for Research in Transportation (hEART).