C2F-TP: A Coarse-to-Fine Denoising Framework for Uncertainty-Aware Trajectory Prediction

Authors: Zichen Wang, Hao Miao, Senzhang Wang, Renzhi Wang, Jianxin Wang, Jian Zhang

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted on two real datasets NGSIM and high D that are widely adopted in trajectory prediction. The result demonstrates the effectiveness of our proposal.
Researcher Affiliation Academia Zichen Wang1, Hao Miao2, Senzhang Wang1*, Renzhi Wang1, Jianxin Wang1, Jian Zhang1 1Central South University 2Aalborg University EMAIL, EMAIL, EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes the methodology using textual explanations and mathematical equations, but it does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/wangzc0422/C2F-TP
Open Datasets Yes The experiments are conducted on two datasets NGSIM (Deo and Trivedi 2018) and high D (Krajewski et al. 2018) that are widely adopted in trajectory prediction.
Dataset Splits Yes We split the dataset into training, validation, and testing sets with a splitting ratio of 7 : 2 : 1.
Hardware Specification Yes We implement our model with the Pytorch framework on a GPU server with NVIDIA 3090 GPU.
Software Dependencies No The paper mentions implementing the model with the 'Pytorch framework' but does not specify a version number or other software dependencies with their versions.
Experiment Setup Yes The parameters in the model are set as follows. We employ a 13 5 grid, which is defined around the target vehicle, where each column corresponds to a single lane, and the rows are separated by a distance of 15 feet. The hidden features of MLP layers are set to 32 with Re Lu as the activation function. To train a coarse-to-fine framework, we consider a two-stage training strategy, where the first stage trains a denoising module and the second stage focuses on training a spatial-temporal interaction module. The details of the two-stage prediction process are given in the associated code repository. Each trajectory is split into segments over a horizon (i.e., 8s), which contains the past (3s) and future (5s) positions at 5Hz.