Drones Help Drones: A Collaborative Framework for Multi-Drone Object Trajectory Prediction and Beyond
Welcome to the official PyTorch implementation of "Drones Help Drones: A Collaborative Framework for Multi-Drone Object Trajectory Prediction and Beyond." We have open-sourced this repository to foster research and collaboration in the field of multi-drone trajectory prediction and related areas.
The implementation code is now available.
This implementation is compatible with PyTorch 2.x, and has been verified to run on NVIDIA H100 GPUs for both training and inference. No additional changes are required—just ensure your CUDA and driver versions support H100, and install the necessary dependencies from environment-torch-2.0.yml.
"Drones Help Drones" has been accepted as a Poster at the Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS 2024). You can access the paper on arXiv.
To set up the environment, use the following command:
conda env create -f environment.ymlEnsure you replace the splits.py file in the nuscenes package (typically found at /miniconda3/envs/dhd/lib/python3.7/site-packages/nuscenes/utils/splits.py) with our provided version of splits.py.
Download the complete Air-Co-Pred dataset, which includes the Trainval dataset (metadata and file blobs parts 0-36), from the following link:
Download Link
Access Code: 4av8
Once downloaded, extract the .tar files into your desired data root directory (YOUR_DATAROOT), organizing them as follows:
Air-Co-Pred/
├── trainval/
│ ├── maps/
│ ├── samples/
│ ├── sweeps/
│ └── v1.0-trainval/To train the DHD (Drones Help Drones) model, execute the following command:
python train.py --config=dhd/config/dhd.yml \
LOG_DIR xxx \
GPUS [x,x,x,x] \
BATCHSIZE 1 \
DATASET.DATAROOT YOUR_DATAROOTTo evaluate the model with trained weights, run:
python test.py --config dhd/config/dhd.yml \
PRETRAINED.LOAD_WEIGHTS True \
PRETRAINED.PATH $YOUR_PRETRAINED_WEIGHTS_PATH \
GPUS [x,x,x,x] \
BATCHSIZE 1 \
DATASET.DATAROOT YOUR_DATAROOTIf you find this work helpful in your research, please consider citing us:
@inproceedings{
title={Drones Help Drones: A Collaborative Framework for Multi-Drone Object Trajectory Prediction and Beyond},
author={Wang Z, Cheng P, Chen M, Tian P, Wang Z, Li X, Yang X, Sun X.},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
year={2024}
}@misc{wang2024droneshelpdronescollaborative,
title={Drones Help Drones: A Collaborative Framework for Multi-Drone Object Trajectory Prediction and Beyond},
author={Zhechao Wang and Peirui Cheng and Mingxin Chen and Pengju Tian and Zhirui Wang and Xinming Li and Xue Yang and Xian Sun},
year={2024},
eprint={2405.14674},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2405.14674},
}