This repository contains the code for the paper "MS-TIP: Imputation Aware Pedestrian Trajectory Prediction," accepted at ICML 2024. MS-TIP is a novel approach designed to enhance pedestrian trajectory prediction by being imputation aware.

Install the dependencies:

conda activate mstip

or

conda env create -f environment.yml --name mstip

To train a MS-TIP model on the dataset, simply run:

./scripts/train.sh -d "eth" -i "0" -n 0.15

• -d flag is the dataset
• -i flag specifies the gpu id
• -n flag specifes how many percentage of data will be set as input during training time. Possible values : 0.05, 0.1, 0.15, 0.2

To evalutate the model performance, simply run:

./scripts/test.sh -d "eth" -i "0" -n 0.15 -b true > output_eth

• The flags have same meaning as above. Set -b flag to true always.

We would like to thank the authors of the following repositories for open-sourcing their code, parts of which were used or referenced in this work: GroupNet, Graph-TERN, PECNet, SAITS.

@inproceedings{chib2024ms,
  title={Ms-tip: Imputation aware pedestrian trajectory prediction},
  author={Chib, Pranav Singh and Nath, Achintya and Kabra, Paritosh and Gupta, Ishu and Singh, Pravendra},
  booktitle={International Conference on Machine Learning},
  pages={8389--8402},
  year={2024},
  organization={PMLR}
}