Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

SAGA and mirror calibration: reproduction data

Data for reproducing Semantic-Geometric Aerial–Ground Registration with Mirror Calibration (CVPR 2027 submission). The code and instructions are in the accompanying code package (see its README.md and docs/DATA.md).

File Size Needed for
saga_results.tar.gz 0.6 GB Level 1: every table, figure and quoted number, from the stored registration outputs
saga_carla_test.tar 13.8 GB Level 2: SAGA re-registration of test-cohort pairs (CARLA observations)
saga_carla_dev.tar 8.5 GB Level 2 on the development cohort (optional)
saga_results_addendum.tar.gz 0.2 GB Level 1 additions for the submitted version: KISS-Matcher without its validity check, graph-only SAGA with ground/non-ground labels, semantic MAC (development), and the inputs of Figures 1 and 2
pip install -U huggingface_hub
hf download FeriSi/saga-cvpr2027-data saga_results.tar.gz --repo-type dataset --local-dir .
tar -xzf saga_results.tar.gz -C workspace      # inside the code package
python reproduce/check_data.py results
hf download FeriSi/saga-cvpr2027-data saga_results_addendum.tar.gz --repo-type dataset --local-dir .
tar -xzf saga_results_addendum.tar.gz -C workspace
python reproduce/check_data.py saga_results_addendum

Each archive unpacks into the code package's workspace/ folder. reproduce/check_data.py verifies every file against the SHA-256 manifests shipped with the code.

Contents

  • CARLA (Town12 and Town13, CARLA 0.9.16): per-attempt retrieval shortlists, registrations of each drone query and its reflection, evaluation labels, true poses, odometry, and per-frame labeled point observations of the drone RGB-D camera and the ground LiDAR.
  • nuScenes (Singapore One-North and Boston): submap and query metadata, ground truth, shortlists, registrations of all methods and references, and the query sweeps used to measure pose error.

License

CC BY-NC-SA 4.0. The nuScenes-derived files are adapted from nuScenes (© Motional, CC BY-NC-SA 4.0) and are shared under the same license for non-commercial research. The CARLA data were generated with the CARLA simulator.

Downloads last month
52