Ours_Waymo_ArmGS preprocessing

Derived preprocessing artifacts used by the ArmGS Waymo experiments released in the model repository with the same name.

License and required notice

This derived dataset was made using the Waymo Open Dataset, provided by Waymo LLC under the Waymo Dataset License Agreement for Non-Commercial Use. Access, use, redistribution, and modification are governed by that agreement, including its non-commercial restrictions.

Read WAYMO_DATASET_LICENSE_NOTICE.md, the included archived agreement, and the current official terms before downloading or using these files.

Important: original sensor data is not included

This repository intentionally excludes original and resized Waymo RGB images, LiDAR records, and raw modular Parquet files. Public redistribution of dataset copies is restricted to registered Waymo users who accepted the Waymo terms. Obtain the source data independently from Waymo and combine it with these derived artifacts.

Excluded local paths include:

• data/waymo_prepared/cache/**/images/**
• data/waymo_prepared/colmap/**/images/**
• data/waymo_prepared/colmap_castrack_centered/**/images/**
• raw Waymo Parquet files

Included artifacts

Under preprocessing/:

• Grounded-SAM sky masks and generation manifests.
• Per-frame actor masks generated from Waymo GT boxes for the ten training-split scenes.
• Known-pose COLMAP databases, dynamic masks, camera models, triangulated binary/text models, and points3D.txt files.
• Centered CAStrack/COLMAP artifacts for the validation reference scene.
• Per-scene preprocessing manifests and CAStrack validation JSON exports.

Absolute paths in copied text manifests were replaced with ${ARMGS_ROOT}, ${WAYMO_ROOT}, or ${WAYMO_V2_ROOT} placeholders. Binary COLMAP databases and models are unchanged.

The ten training scenes have complete RGB-index manifests, sky masks, actor masks, and COLMAP outputs. The RGB counts are 197--199 per scene, the held-out evaluation count is 49, and the known-pose COLMAP training count is 148--150. See release_manifest.json for exact per-scene counts.

The training-split preprocessing uses Waymo GT lidar_box actor tracking fallback, not CAStrack, and corresponds to streetgs-periodic rather than official SplatAD LINSPACE50.

Download

hf download Dororo99/Ours_Waymo_ArmGS \
  --repo-type dataset \
  --local-dir Ours_Waymo_ArmGS_preprocessing

Citation

@misc{waymo_open_dataset,
  title   = {Waymo Open Dataset: An autonomous driving dataset},
  website = {https://www.waymo.com/open},
  year    = {2019--2025}
}