G2LTraj: A Global-to-Local Generation Approach for Trajectory Prediction

Authors: Zhanwei Zhang, Zishuo Hua, Minghao Chen, Wei Lu, Binbin Lin, Deng Cai, Wenxiao Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our G2LTraj significantly improves the performance of seven existing trajectory predictors across the ETH, UCY and nu Scenes datasets. Experimental results demonstrate its effectiveness.
Researcher Affiliation Collaboration Zhanwei Zhang1 , Zishuo Hua2 , Minghao Chen3 , Wei Lu5 , Binbin Lin2,4 , Deng Cai1 and Wenxiao Wang2 1State Key Lab of CAD&CG, Zhejiang University 2School of Software Technology, Zhejiang University 3Hangzhou Dianzi University 4Zhiyuan Research Institute 5Ningbo Beilun Third Container Terminal Co., Ltd
Pseudocode Yes Algorithm 1 Global-to-local Generation Process Input: The granularity of key steps L, the predicted key coordinates { ˆZ1+i L}N i=0, two-layer MLPs {ϕl}l {L,L/2,...,2}, the position embedding {p }1+NL =0 , and agent features A. Output: A predicted future trajectory ˆFL R(1+NL) 2 for agent a. 1: Let l L. 2: while l > 1 do 3: Calculate n ˆZ1+(i+ 1 i=0 by Eq. (6) in parallel with ϕl, ˆZ1+il, p1+il, ˆZ1+(i+1)l, p1+(i+1)l and A as input; 4: l l/2. 5: l denotes the current interval of existing adjacent predicted steps. Notably, we set L {2, 4, 8, ...} to ensure l/2 is an integer value continuously. When l = 1, it indicates that the entire trajectory coordinates of 1 + NL steps have already been generated. 6: end while
Open Source Code Yes Code will be available at https://github. com/Zhanwei-Z/G2LTraj.
Open Datasets Yes Datasets. We conducted experiments on three widely-used datasets: ETH [Pellegrini et al., 2009], UCY [Lerner et al., 2007] and nu Scenes [Caesar et al., 2020].
Dataset Splits Yes We employ the standard leave-one-out strategy [Bae et al., 2023; Alahi et al., 2016] for training and evaluation. The task in ETH and UCY is to predict the future 12 steps (4.8 seconds) based on the observed 8 steps (3.2 seconds). nu Scenes is a large-scale trajectory dataset that includes both vehicles and pedestrians. The task in nu Scenes is to predict the future 12 steps (6 seconds) given the observed 4 steps (2 seconds) and the High-Definition (HD) maps. We evaluate the results on nu Scenes validation set with K=10, following [Girgis et al., 2022].
Hardware Specification Yes The model is trained using a single 3090 GPU.
Software Dependencies No The paper mentions optimizers (Adam W, Adam) and neural network components (MLP), but it does not specify version numbers for any programming languages, frameworks, or libraries used in the implementation.
Experiment Setup Yes Specifically, for ETH and UCY, we employ the Adam W optimizer [Loshchilov and Hutter, 2018], set the batch size to 128, use a learning rate of 0.001, and train for 256 epochs. For nu Scenes, we utilize the Adam optimizer [Kingma and Ba, 2014], set the batch size to 64, use a learning rate of 7.5e 4, and train for 80 epochs.