MichaelH717 / ADAPT_MH

Official implementation of the ICCV 2023 paper "ADAPT: Efficient Multi-Agent Trajectory Prediction with Adaptation"

ADAPT: Efficient Multi-Agent Trajectory Prediction with Adaptation

Paper | Webpage

This is the official implementation of the paper ADAPT: Efficient Multi-Agent Trajectory Prediction with Adaptation published in ICCV 2023.

Introduction

1. Clone this repository:

clone https://github.com/gorkaydemir/ADAPT.git
cd ADAPT

2. Create a conda environment and install required packages:

conda create -n adapt python=3.8
conda activate adapt
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch

3. Apply preprocessing to data and extract them into /path/to/data, following the process explained in dataset/README.md

Train

python run.py \
--ex_file_path /path/to/data/extended_ex_list \
--val_ex_file_path /path/to/data/eval.ex_list \
--model_save_path checkpoints/exp0 \
--static_agent_drop --scaling

Validation

python run.py --validate \
--ex_file_path /path/to/data/extended_ex_list \
--val_ex_file_path /path/to/data/eval.ex_list \
--model_save_path checkpoints/exp0 \
--checkpoint_path /path/to/checkpoint --use_checkpoint 

You can download the pretrained model here.

How to Cite

@InProceedings{Aydemir2023ICCV,
        author = {Aydemir, G\"orkay and Akan, Adil Kaan and G\"uney, Fatma},
        title = {{ADAPT}: Efficient Multi-Agent Trajectory Prediction with Adaptation},
        booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision},
        year      = {2023}}