We implemented SurroundDepth as the baseline model for Track 4
. The baseline model was trained on the official train
split of the nuScenes dataset and evaluated on our robustness probing sets under different corruptions.
This codebase provides basic instructions for the reproduction of the baseline model in the RoboDrive Challenge.
Kindly refer to README.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 set. For training data preparation, kindly refer to README.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
│ │ ├── depth
│ │ └── raw_data
│ └── robodrive
├── datasets
│ ├── __init__.py
│ ├── corruption_dataset.py
│ ├── ddad_dataset.py
│ ├── mono_dataset.py
│ ├── nusc_dataset.py
│ ├── nusc
│ ├── ddad
│ └── robodrive
│
...
Then copy the ./robodrive/robodrive-v1.0-test
to the nuscenes folder by running:
cp -r ./data/robodrive/robodrive-v1.0-test ./data/nuscenes/raw_data
The training and evaluation instructions are summarized as follows.
Kindly refer to README.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 SurroundDepth
bash eval.sh
The generated results will be saved in the folder structure as follows. Each results.pkl
is a dictionary, its key is sample_idx
and its value is np.ndarray
.
.
├── brightness
│ └── results.pkl
├── color_quant
│ └── results.pkl
├── contrast
│ └── results.pkl
...
├── snow
└── zoom_blur
Next, kindly merge all the .pkl
files into a single pred.npz
file.
You can merge the results using the following command:
python ./convert_submit.py
⚠️ Note: The prediction file MUST be named aspred.npz
.
Finally, upload the compressed file to Track 4
'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 CorruptionDataset
from this link.
We mainly modified the data loading part. We only consider the subset of scenes for each corruption type, below is an example showing how to load a subset of scenes under each corruption type.
cam_sample = self.nusc.get(
'sample_data', rec['data'][self.camera_names[index_spatial]])
inputs['id'].append(self.camera_ids[index_spatial])
# modify the data path to corruption data path here
color = self.loader(os.path.join(self.corruption_root, self.corruption, cam_sample['filename']))
Metric | Results |
---|---|
Abs Rel | 0.348 |
Sq Rel | 4.336 |
RMSE | 7.102 |
RMSE Log | 0.396 |
a1 | 0.623 |
a2 | 0.819 |
a3 | 0.899 |
Metric | Results |
---|---|
Abs Rel | 0.304 |
Sq Rel | 3.060 |
RMSE | 8.527 |
RMSE Log | 0.400 |
a1 | 0.544 |
a2 | 0.784 |
a3 | 0.891 |
Kindly cite the corresponding paper(s) once you use the baseline model in this track.
@inproceedings{wei2023surrounddepth,
title={SurroundDepth: Entangling Surrounding Views for Self-Supervised Multi-Camera Depth Estimation},
author={Wei, Yi and Zhao, Linqing and Zheng, Wenzhao and Zhu, Zheng and Rao, Yongming and Huang, Guan and Lu, Jiwen and Zhou, Jie},
booktitle={Conference on Robot Learning},
pages={539-549},
year={2023},
organization={PMLR}
}