Datasets:
Tasks:
Image-to-3D
Modalities:
Geospatial
Languages:
English
Size:
100K<n<1M
Tags:
3d-point-cloud
point-cloud-generation
city-scale
remote-sensing
satellite-imagery
digital-surface-model
License:
| #!/usr/bin/env python3 | |
| """ | |
| City3D-MultiGen — pipeline runner. | |
| This runs the reconstruction stages for one city in order. It does NOT host any | |
| data: it drives the same scripts documented in the README to rebuild the aligned | |
| multi-modal tiles locally from (1) source 3D data you downloaded yourself and | |
| (2) live Google Maps Static API calls made under your own key. | |
| Manual prerequisites (NOT automated here — see README.md): | |
| 1. Download the source 3D data: | |
| - Melbourne: City of Melbourne 3D Point Cloud 2018 (LAS). | |
| - HoliCity (London): FBX meshes, then sample a point cloud to LAS with | |
| CloudCompare, and georeference it with holicity/convert_coord.py and | |
| holicity/add_coord_head.py. | |
| 2. Install PDAL (conda install -c conda-forge pdal) — the tiler calls it. | |
| 3. Export your Google credentials: | |
| GOOGLE_MAPS_API_KEY, GOOGLE_MAPS_URL_SIGNING_SECRET, GOOGLE_MAPS_STYLE_MAP_ID | |
| 4. Set the input/output paths at the top of each stage script (the tilers read | |
| their LAS input dir and output dir from module-level constants). | |
| Stages run by this script (per city): | |
| A. <city>/export_las_blocks_noKML.py -> tiles + per-tile DSM + BEV | |
| B. <city>/Obtain_corresponding_map_signed.py -> satellite + 6 semantic masks | |
| C. make_splits.py -> train/val/test tile lists | |
| Usage: | |
| python scripts/build_dataset.py --city melbourne | |
| python scripts/build_dataset.py --city holicity --data_root ./output --skip_splits | |
| """ | |
| import argparse | |
| import os | |
| import shutil | |
| import subprocess | |
| import sys | |
| HERE = os.path.dirname(os.path.abspath(__file__)) | |
| ENV_VARS = ("GOOGLE_MAPS_API_KEY", "GOOGLE_MAPS_URL_SIGNING_SECRET", "GOOGLE_MAPS_STYLE_MAP_ID") | |
| def check_prereqs(): | |
| missing = [k for k in ENV_VARS if not os.environ.get(k)] | |
| if missing: | |
| sys.exit(f"[error] Missing environment variables: {', '.join(missing)}. See README.md.") | |
| if shutil.which("pdal") is None: | |
| sys.exit("[error] PDAL not found on PATH. Install it: conda install -c conda-forge pdal") | |
| def run(script_rel, *cli_args): | |
| path = os.path.join(HERE, script_rel) | |
| cmd = [sys.executable, path, *cli_args] | |
| print(f"\n>>> {' '.join(cmd)}", flush=True) | |
| subprocess.run(cmd, check=True) | |
| def main(): | |
| ap = argparse.ArgumentParser(description="Run the City3D-MultiGen reconstruction stages.") | |
| ap.add_argument("--city", choices=["melbourne", "holicity"], required=True) | |
| ap.add_argument("--data_root", default="./output", | |
| help="Directory holding the assembled tiles (used for the split step).") | |
| ap.add_argument("--skip_splits", action="store_true", help="Do not run make_splits.py.") | |
| args = ap.parse_args() | |
| check_prereqs() | |
| print(f"[info] Running the {args.city} pipeline. Ensure the manual prerequisites in this " | |
| f"script's docstring are done and paths are configured at the top of each stage script.") | |
| # Stage A — tile the (already downloaded / sampled) source point clouds. | |
| run(f"{args.city}/export_las_blocks_noKML.py") | |
| # Stage B — fetch satellite + semantic maps for the tiles produced in Stage A. | |
| if args.city == "melbourne": | |
| run("melbourne/Obtain_corresponding_map_signed.py", "--folder", args.data_root) | |
| else: | |
| run("holicity/Obtain_corresponding_map_signed.py") | |
| # Stage C — generate the train/val/test split lists. | |
| if not args.skip_splits: | |
| run("make_splits.py", "--data_root", args.data_root, | |
| "--out_dir", os.path.join(HERE, "..", "metadata", "splits")) | |
| print("\n[done] Reminder: Google Maps imagery is subject to the Google Maps Platform ToS; " | |
| "do not redistribute the fetched *_sat.png / *_map.png / *_<Class>.png files.") | |
| if __name__ == "__main__": | |
| main() | |