envship_v2_datasets / scripts /extras /pbf_to_tile_cache.py
mark000071's picture
v2.1 super-squash
6cbe918
Raw
History Blame Contribute Delete
7.11 kB
#!/usr/bin/env python3
"""Convert a Geofabrik OSM PBF extract into the per-tile Overpass-JSON
cache format expected by build_standard_track_context_v1 (used inside the EnvShip-Bench build pipeline).
For each tile (0.25 degree by default), emit a JSON file at:
<out_cache>/tiles/<TILE_ID>.json
where TILE_ID matches the convention used by the build script:
f"{tile_lat:+08.3f}_{tile_lon:+09.3f}"
The JSON structure mirrors the Overpass API output that the build script's
_parse_ways() consumes — list of {"type": "way", "id": ..., "nodes": [...],
"tags": {"natural":"coastline"} | {"natural":"water"} |
{"man_made":"pier"|"breakwater"|"groyne"|"quay"}, "geometry": [{"lat":..,
"lon":..}, ...]} entries inside payload["elements"].
This script extracts only the way geometry types that the build script
queries via Overpass — keeping the cache size small and the parse fast.
Usage
-----
python pbf_to_tile_cache.py \
--pbf <path.osm.pbf> \
--out-cache <env/osm_cache or alt path> \
--bbox south,west,north,east \
[--tiles-list <file with one TILE_ID per line>] \
[--tile-deg 0.25]
Either --bbox (full coverage) or --tiles-list (selective) MUST be given.
"""
from __future__ import annotations
import argparse
import json
import math
import os
import sys
import time
from collections import defaultdict
from pathlib import Path
from typing import Iterable
import osmium
def tile_id_for_point(lat: float, lon: float, tile_deg: float) -> str:
tile_lat = math.floor(lat / tile_deg) * tile_deg
tile_lon = math.floor(lon / tile_deg) * tile_deg
return f"{tile_lat:+08.3f}_{tile_lon:+09.3f}"
# Tags accepted, matching _make_overpass_query() in build_standard_track_context_v1 (used inside the EnvShip-Bench build pipeline)
def is_target_way(tags: dict) -> bool:
if not tags:
return False
nat = tags.get("natural", "")
mm = tags.get("man_made", "")
if nat == "coastline":
return True
if nat == "water":
return True
if mm in ("pier", "breakwater", "groyne", "quay"):
return True
return False
class WayCollector(osmium.SimpleHandler):
"""Collect target ways with geometry into per-tile buckets.
osmium gives us node coordinates by attaching a location cache before
parsing. The SimpleHandler base class supports `apply_file(filename,
locations=True)` which fills the cache automatically.
"""
def __init__(self, tile_deg: float, accept_tiles: set | None = None,
bbox: tuple | None = None):
super().__init__()
self.tile_deg = tile_deg
self.accept_tiles = accept_tiles
self.bbox = bbox # (south, west, north, east) or None
self.tile_payload: dict[str, list[dict]] = defaultdict(list)
self.way_count = 0
self.kept_count = 0
self.t0 = time.time()
def way(self, w):
self.way_count += 1
if self.way_count % 100000 == 0:
sys.stderr.write(
f"[pbf] ways scanned={self.way_count:,} kept={self.kept_count:,} "
f"elapsed={time.time()-self.t0:.0f}s\n")
tags = {tag.k: tag.v for tag in w.tags}
if not is_target_way(tags):
return
try:
coords = []
for n in w.nodes:
# osmium WayNode -> location
loc = n.location
if loc.valid():
coords.append((loc.lat, loc.lon))
except osmium.InvalidLocationError:
return # unresolved node coords; skip
if len(coords) < 2:
return
# bbox filter
if self.bbox is not None:
s, ww, n, ee = self.bbox
lats = [c[0] for c in coords]; lons = [c[1] for c in coords]
if max(lats) < s or min(lats) > n or max(lons) < ww or min(lons) > ee:
return
# Group by tile: assign way to every tile that any node falls in
node_tiles = set()
for lat, lon in coords:
node_tiles.add(tile_id_for_point(lat, lon, self.tile_deg))
if self.accept_tiles is not None:
node_tiles &= self.accept_tiles
if not node_tiles:
return
elem = {
"type": "way",
"id": w.id,
"tags": tags,
"geometry": [{"lat": float(lat), "lon": float(lon)} for lat, lon in coords],
}
for tid in node_tiles:
self.tile_payload[tid].append(elem)
self.kept_count += 1
def write_tile_jsons(payload_map: dict[str, list[dict]], out_root: Path) -> int:
tiles_dir = out_root / "tiles"
tiles_dir.mkdir(parents=True, exist_ok=True)
n = 0
for tid, elements in payload_map.items():
path = tiles_dir / f"{tid}.json"
payload = {
"version": 0.6,
"generator": "pbf_to_tile_cache.py",
"osm3s": {
"copyright":
"The data included in this document is from www.openstreetmap.org. "
"Available under the Open Database License (ODbL).",
},
"elements": elements,
}
path.write_text(json.dumps(payload))
n += 1
return n
def main():
p = argparse.ArgumentParser()
p.add_argument("--pbf", type=Path, required=True)
p.add_argument("--out-cache", type=Path, required=True,
help="Destination root; tiles/ subdir will be created.")
p.add_argument("--bbox", type=str, default=None,
help="south,west,north,east — restrict to this area.")
p.add_argument("--tiles-list", type=Path, default=None,
help="One TILE_ID per line; only emit these tiles. "
"If absent emit all tiles intersected by ways within --bbox.")
p.add_argument("--tile-deg", type=float, default=0.25)
args = p.parse_args()
accept_tiles = None
if args.tiles_list:
accept_tiles = {ln.strip() for ln in args.tiles_list.read_text().splitlines() if ln.strip()}
print(f"[pbf] accept_tiles loaded: {len(accept_tiles):,}", flush=True)
bbox = None
if args.bbox:
s, ww, n, ee = [float(x) for x in args.bbox.split(",")]
bbox = (s, ww, n, ee)
print(f"[pbf] bbox=({s},{ww},{n},{ee})", flush=True)
coll = WayCollector(args.tile_deg, accept_tiles=accept_tiles, bbox=bbox)
print(f"[pbf] parsing {args.pbf} size={args.pbf.stat().st_size/(1<<20):.0f} MiB", flush=True)
# locations=True attaches the node location cache (in-RAM dense index by id)
# idx="sparse_mem_array" is more compact for partial PBFs; use sparse_mmap_array
# for big planet files.
coll.apply_file(str(args.pbf), locations=True, idx="sparse_mem_array")
print(f"[pbf] DONE ways={coll.way_count:,} kept={coll.kept_count:,} "
f"tiles_with_data={len(coll.tile_payload):,}", flush=True)
written = write_tile_jsons(coll.tile_payload, args.out_cache)
print(f"[pbf] wrote {written:,} tile JSON files under {args.out_cache}/tiles/", flush=True)
if __name__ == "__main__":
main()