Den-TP: A Density-Balanced Data Curation and Evaluation Framework for Trajectory Prediction

This repository contains the official implementation of Den-TP: A Density-Balanced Data Curation and Evaluation Framework for Trajectory Prediction, an efficient framework for constructing compact and balanced training subsets for trajectory prediction.

Getting Started

Step 1: Clone the repo

git clone https://github.com/<anonymized-repo-link>.git
cd SSTP

Step 2: Set up the environment

conda env create -f SSTP.yml
conda activate SSTP

Step 3: Download data

Download the Argoverse Motion Forecasting Dataset v1.1 and prepare the directory following the official dataset structure.

Step 4: Install the Argoverse API

Follow the official Argoverse API installation guide.

Select Sample

To run data selection with customized settings:

bash select.sh