Safety-aware Motion Prediction with Unseen Vehicles for Autonomous Driving
Xuanchi Ren, Tao Yang, Li Erran Li, Alexandre Alahi, and Qifeng Chen
ICCV 2021
[Paper] [Supplementary material]
✅ Update data preprocessing code
✅ Update model
🔲 Training script
- Clone this repository with the following command:
git clone https://github.com/xrenaa/Safety-Aware-Motion-Prediction.git
cd experiments/nuScenes
git clone https://github.com/nutonomy/nuscenes-devkit.git
git checkout 12fb09169eb8ebf04bc39a30cd50334215769c3e
- Replace
experiments/nuScenes/nuscenes-devkit/python-sdk/nuscenes/prediction/input_representation/static_layers.pywith the file Here.
First, we'll create a conda environment to hold the dependencies.
conda create --name safeDrive python=3.6 -y
source activate safeDrive
pip install -r requirements.txt
- Download the nuScenes dataset (this requires signing up on their website). Extract the downloaded zip file's contents and place them in the
experiments/nuScenesdirectory. Then, download the map expansion pack (v1.2) and copy the contents of the extractedmapsfolder into theexperiments/nuScenes/mapsfolder. Eventually you should have the following folder structure:
experiments/nuscenes
samples - Sensor data for keyframes.
sweeps - Sensor data for intermediate frames.
maps - Folder for all map files: rasterized .png images and vectorized .json files.
v1.0-trainval - JSON tables that include all the meta data and annotations.
process_data.py - Our provided data processing script.
- Finally, process them into a data format that our model can work with.
cd experiments/nuScenes
# For the tranval nuScenes dataset, use the following
python process_data.py --data ../nuScenes --split train --img_size 128
python process_data.py --data ../nuScenes --split train_val --img_size 128
python process_data.py --data ../nuScenes --split val --img_size 128
We provide a notebook to visualize the processed data.
@inproceedings{ren2021unseen,
title = {Safety-aware Motion Prediction with Unseen Vehicles for Autonomous Driving},
author = {Ren, Xuanchi, and Yang, Tao, and Li, Li Erran, and Alahi, Alexandre, and Chen, Qifeng},
booktitle = {ICCV},
year = {2021}
}