We implemented BEVFusion as the baseline model for Track 5
. The baseline model was trained on the official train
split of the nuScenes dataset and evaluated on our robustness probing sets under sensor failure scenarios.
This codebase provides basic instructions for the reproduction of the baseline model in the RoboDrive Challenge.
Kindly refer to GET_STARTED.md to set up environments and download necessary checkpoints.
We use data under the nuScenes train
split as the training set and the RoboDrive robustness probing data as the evaluation sets. For training data preparation, kindly refer to NUSCENES_DET.md.
For evaluation data preparation, kindly download the dataset from the following resources:
Type | Phase 1 | Phase 2 |
---|---|---|
Google Drive | link1 or link2 |
link1 or link2 |
Uncompress the downloaded dataset and organize the folder structure as follows:
.
├── data
│ ├── nuscenes
│ └── robodrive-sensor
├── configs
├── mmdet3d
└── tools
Next, run the following command to generate the .pkl
file for the evaluation sets:
bash tools/create_data.sh
🚙 Hint: You can download our generated
.pkl
file from this Google Drive link.
The nuscenes
folder should end up looking like this:nes folder should be like this:
.
├── basemap
├── can_bus
├── can_bus.zip
├── expansion
├── lidarseg
├── maps
├── nuscenes_infos_train.pkl
├── nuscenes_infos_val.pkl
├── nuScenes-panoptic-v1.0-all
├── prediction
├── robodrive_infos_test.pkl
├── robodrive-v1.0-test
├── samples
├── sweeps
├── v1.0-mini
├── v1.0-test
└── v1.0-trainval
The training and evaluation instructions are summarized as follows.
Kindly refer to GET_STARTED.md for the details regarding model training.
Simply run the following command to evaluate the trained baseline model on the RoboDrive robustness probing sets:
cd BEVFusion
bash tools/test_corruption.sh
Please rename the generated results_nusc.json
file to pred.json
and compress it into a .zip
file.
Finally, upload the compressed file to Track 5
's evaluation server for model evaluation.
🚙 Hint: We provided the baseline submission file at this Google Drive link. Feel free to download and check it for reference and learn how to correctly submit the prediction files to the server.
To customize your own dataset, simply build your dataset based on RoboDriveDataset
from this line. We simply modified the data path to load the image and LiDAR data.
Model | NDS | mAP | mATE | mASE | mAOE | mAVE | mAAE |
---|---|---|---|---|---|---|---|
BEVFusion | 0.4285 | 0.2448 | 0.4012 | 0.2910 | 0.4928 | 0.5289 | 0.2251 |
Model | NDS | mAP | mATE | mASE | mAOE | mAVE | mAAE |
---|---|---|---|---|---|---|---|
BEVFusion | 0.3913 | 0.2159 | 0.4402 | 0.2996 | 0.5297 | 0.6462 | 0.2506 |
Kindly cite the corresponding paper(s) once you use the baseline model in this track.
@inproceedings{liu2023bevfusion,
title = {BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's Eye View Representation},
author = {Liu, Zhijian and Tang, Haotian and Amini, Alexander and Yang, Xinyu and Mao, Huizi and Rus, Daniela L and Han, Song},
booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
pages = {2774-2781},
year = {2023}
}