SkyScenes Dataset

Description

Pipeline for creating synthetic aerial and ground-level dataset for visual odometry and depth estimation.

Data Generation

The project uses a unified pipeline for data generation.

1. Configure: Edit config/default_generation.yaml to set your desired towns, weather, and camera parameters.
2. Run:

   uv run python pipeline/run_generation.py

   By default, this generates RGB images and pose metadata, skipping heavy segmentation/depth maps to save space. To enable all ground truth maps, set save_seg: true in the config or use --save_seg flag.

Pose Conversion

To convert generated camera poses to KITTI format:

uv run python scripts/convert_poses_to_kitti.py --input_dir /home/df/data/datasets/H_35_P_45/ClearNoon/Town01/metaData --output_file poses.txt

NOTE

Data should be stored in following directory: /home/df/data/datasets