South Dataset Preparation
Terms of Use
By accessing or using this dataset, you agree to the following terms:
- The dataset must not be distributed, shared, sublicensed, or redistributed, in whole or in part, without prior written permission from the dataset owners.
- The dataset may be used for research purposes only.
This directory contains the assets and preprocessing script for preparing the South dataset.
Directory Structure
Place the South dataset under data/south/ with the following layout:
data/
└── south/
├── left_images/
├── sat_images/
├── depths/
├── left_camera_poses.txt
├── sat_camera_poses.txt
├── pairs.txt
└── process_data.py
left_images/: left camera images.sat_images/: satellite BEV images.depths/: depth maps corresponding to the left camera images.left_camera_poses.txt,sat_camera_poses.txt, andpairs.txt: metadata used by the preprocessing script.
Preprocess the Dataset
From the project root, run:
cd data/south
python process_data.py
The script generates south_data.pth in the same directory.
Validate with LuSNAR
To conduct zero-shot testing on the South dataset after training on the LuSNAR dataset, copy the processed South dataset file into data/LuSNAR/:
cp data/south/south_data.pth data/LuSNAR/
Then update configs/warg_configs/warg_config.yaml to include the South split during testing:
data_keys: [train, val, test, south]
Finally, run training:
python train.py
- Downloads last month
- 10