MGF: Mixed Gaussian Flow for Diverse Trajectory Prediction

Authors: Jiahe Chen, Jinkun Cao, Dahua Lin, Kris Kitani, Jiangmiao Pang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental It achieves state-of-the-art performance in the evaluation of both trajectory alignment and diversity on the popular UCY/ETH and SDD datasets.
Researcher Affiliation Collaboration Jiahe Chen1,2 Jinkun Cao3 Dahua Lin4,2,5 Kris Kitani3 Jiangmiao Pang2 1 Zhejiang University 2Shanghai AI Laboratory 3Carnegie Mellon University 4The Chinese University of Hong Kong 5CPII under Inno HK
Pseudocode No The paper describes the forward and inverse processes of its model in text format with equations but does not present a formal pseudocode block or algorithm chart.
Open Source Code Yes Code is available at https://github.com/mulplue/MGF.
Open Datasets Yes We evaluate on two major benchmarks, i.e., ETH/UCY [24, 38] and SDD [42].
Dataset Splits Yes Social-GAN follows a specific scheme and ratio for splitting the data into train/val/test sets.
Hardware Specification Yes All models were trained on a single NVIDIA V100 GPU for 100 epochs(approximately 4 to 8 hours).
Software Dependencies No The paper mentions using components like Real NVP and CIFs from other works but does not list specific version numbers for software dependencies like Python, PyTorch, or other libraries used in the environment.
Experiment Setup Yes We enhance our model using a similar technique as intension clustering" and we name it prediction clustering". ... We follow Flow Chain s implementations of CIFs that each layer consists of a Real NVP [10] with a 3-layer MLP and 128 hidden units. We use a Trajectron++ [46] encoder to encode historical trajectories. ... By default, we perform clustering by K-means with K = 8.