File size: 8,946 Bytes
86c018f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 | """
ObjectPropertiesProcessor — 3dSAGER 25 geometric property computation.
Given a dict of {building_id: mesh_record} for each side (cands/index),
computes exactly the 25 properties used in the SIGMOD 2026 paper and
applies log(1+x) normalisation if requested.
mesh_record format (from crawl_3dbag_multilod.py):
{'polygon_mesh': [[[x,y,z],...], ...], # list of surfaces
'vertices': np.array unique vertices,
'centroid': np.array([x,y,z])}
"""
import time, numpy as np
from scipy.spatial import ConvexHull
PROP_NAMES = [
"bounding_box_width", "bounding_box_length", "area", "perimeter",
"perimeter_ind", "volume", "convex_hull_area", "convex_hull_volume",
"ave_centroid_distance", "height_diff", "num_floors", "axes_symmetry",
"compactness_2d", "compactness_3d", "density", "elongation",
"shape_ind", "hemisphericality", "fractality", "cubeness",
"circumference", "aligned_bounding_box_width",
"aligned_bounding_box_length", "aligned_bounding_box_height",
"num_vertices"
]
def _extract_faces(mesh_rec):
"""Return (list-of-face-vertex-arrays, all-vertices-array)."""
pm = mesh_rec.get("polygon_mesh", [])
faces = []
all_pts = []
for surf in pm:
if len(surf) < 3:
continue
verts = np.asarray(surf, dtype=np.float64)
faces.append(verts)
all_pts.append(verts)
if not faces:
all_v = mesh_rec.get("vertices", np.zeros((0, 3)))
return [], np.asarray(all_v, dtype=np.float64)
all_v = np.concatenate(all_pts, axis=0)
return faces, all_v
def _footprint_vertices(all_v):
"""Return 2D vertices (x,y only) for the building footprint."""
if len(all_v) < 3:
return all_v[:, :2]
return all_v[:, :2]
def _unique_2d(pts):
"""Deduplicate 2D points."""
if len(pts) <= 1:
return pts
_, idx = np.unique(np.round(pts, 6), axis=0, return_index=True)
return pts[np.sort(idx)]
def _triangle_area(a, b, c):
"""Area of triangle a-b-c."""
return 0.5 * np.linalg.norm(np.cross(b - a, c - a))
def _face_area(verts):
"""Area of a planar polygon (triangulation from first vertex)."""
if len(verts) < 3:
return 0.0
a = 0.0
p0 = verts[0]
for j in range(1, len(verts) - 1):
a += _triangle_area(p0, verts[j], verts[j + 1])
return a
def _face_perimeter(verts):
"""Perimeter of a polygon."""
if len(verts) < 2:
return 0.0
p = 0.0
for j in range(len(verts)):
p += np.linalg.norm(verts[j] - verts[(j + 1) % len(verts)])
return p
def _mesh_volume(faces):
"""Signed volume via divergence theorem: V = 1/3 * sum over faces
of (face centroid · face normal) * face area. Works for closed meshes."""
vol = 0.0
for fv in faces:
if len(fv) < 3:
continue
p0 = fv[0]
c = fv.mean(axis=0)
for j in range(1, len(fv) - 1):
n = np.cross(fv[j] - p0, fv[j + 1] - p0)
vol += np.dot(c, n)
return abs(vol) / 6.0
def _convex_hull_volume(all_v):
"""Volume of 3D convex hull."""
if len(all_v) < 4:
return 0.0
try:
hull = ConvexHull(all_v)
return hull.volume
except Exception:
return 0.0
def _convex_hull_area_2d(pts_2d):
"""Area of 2D convex hull (footprint)."""
pts_u = _unique_2d(pts_2d)
if len(pts_u) < 3:
return np.ptp(pts_u[:, 0]) * np.ptp(pts_u[:, 1]) if len(pts_u) > 1 else 0.0
try:
hull = ConvexHull(pts_u)
return hull.volume # in 2D, hull.volume = area
except Exception:
return 0.0
def _footprint_perimeter_and_area(pts_2d):
"""Footprint (2D convex hull) perimeter and area."""
pts_u = _unique_2d(pts_2d)
if len(pts_u) < 3:
return 0.0, 0.0
try:
hull = ConvexHull(pts_u)
return hull.area, hull.volume # perimeter = hull.area, area = hull.volume
except Exception:
return 0.0, 0.0
def _pca_axes(all_v):
"""Return sorted eigenvalues (variances along principal axes)."""
if len(all_v) < 3:
return np.array([1.0, 1.0, 1.0])
c = all_v.mean(axis=0)
X = all_v - c
cov = X.T @ X / (len(X) - 1 + 1e-12)
try:
w, _ = np.linalg.eigh(cov)
return np.sort(np.maximum(w, 0))[::-1]
except Exception:
return np.array([1.0, 1.0, 1.0])
def _compute_all(all_v, faces, mesh_rec):
"""Compute all 25 property values from vertices and faces."""
eps = 1e-9
props = {}
# bbox in world coords
if len(all_v) == 0:
return {p: 0.0 for p in PROP_NAMES}
bmin, bmax = all_v.min(axis=0), all_v.max(axis=0)
bbox_dims = bmax - bmin # [width_x, width_y, height]
bbw, bbl, bbh = float(bbox_dims[0]), float(bbox_dims[1]), float(bbox_dims[2])
props["bounding_box_width"] = max(bbw, eps)
props["bounding_box_length"] = max(bbl, eps)
props["aligned_bounding_box_width"] = max(bbw, eps)
props["aligned_bounding_box_length"] = max(bbl, eps)
props["aligned_bounding_box_height"] = max(bbh, eps)
props["height_diff"] = max(bbh, eps)
props["num_floors"] = max(bbh / 3.0, eps) # ~3m per floor
# area = sum of face areas
area = sum(_face_area(f) for f in faces)
props["area"] = max(area, eps)
# perimeter = sum of face perimeters
perim = sum(_face_perimeter(f) for f in faces)
props["perimeter"] = max(perim, eps)
# volume via divergence theorem
vol = _mesh_volume(faces)
props["volume"] = max(vol, eps)
# convex hull
pts_2d = _footprint_vertices(all_v)
props["convex_hull_area"] = max(_convex_hull_area_2d(pts_2d), eps)
props["convex_hull_volume"] = max(_convex_hull_volume(all_v), eps)
# footprint perimeter & area (for compactness / perimeter_ind)
fp_perim, fp_area = _footprint_perimeter_and_area(pts_2d)
props["perimeter_ind"] = max(fp_perim / (2.0 * np.sqrt(np.pi * max(fp_area, eps)) + eps), eps)
props["circumference"] = max(fp_perim, eps)
# centroid distance
centroid = np.asarray(mesh_rec.get("centroid", all_v.mean(axis=0)), dtype=np.float64)
dists = np.linalg.norm(all_v - centroid, axis=1)
props["ave_centroid_distance"] = max(float(dists.mean()), eps)
# PCA axes
axes = _pca_axes(all_v) # sorted descending
props["elongation"] = max(float(axes[0] / (axes[2] + eps)), eps)
props["axes_symmetry"] = max(float((axes[1] + axes[2]) / (2.0 * max(axes[0], eps))), eps)
# compactness
props["compactness_2d"] = max(float(4.0 * np.pi * max(fp_area, eps) / (max(fp_perim, eps) ** 2 + eps)), eps)
# 3D compactness: V^2 / A^3 * 36π (sphere = 1)
area_safe = max(area, eps)
vol_safe = max(vol, eps)
props["compactness_3d"] = max(float(36.0 * np.pi * vol_safe * vol_safe / (area_safe * area_safe * area_safe + eps)), eps)
# density
props["density"] = max(float(vol_safe / max(fp_area, eps)), eps)
# shape index
props["shape_ind"] = max(float(bbh / (np.sqrt(max(fp_area, eps)) + eps)), eps)
# hemisphericality: how close to a hemisphere (min height vs max height)
props["hemisphericality"] = max(float(1.0 - (all_v[:, 2].min() - bmin[2]) / (max(bbh, eps))), eps)
# fractality: log(area) / log(perimeter) as a rough fractal dimension proxy
props["fractality"] = max(float(np.log(max(area, 1.0)) / np.log(max(perim, 1.0))), eps)
# cubeness: V / bbox_volume
bbox_vol = max(bbw * bbl * bbh, eps)
props["cubeness"] = max(float(vol_safe / bbox_vol), eps)
# num_vertices
props["num_vertices"] = max(len(all_v), eps)
return props
class ObjectPropertiesProcessor:
def __init__(self, object_dict, vector_normalization=True):
self.od = object_dict
self.vector_normalization = vector_normalization
self.property_dict_generation_time = 0.0
t0 = time.time()
self.prop_vals_dict = self._compute()
self.property_dict_generation_time = time.time() - t0
def _resolve_sides(self):
"""Object dict may use {'cands':..., 'index':...} or arbitrary side keys.
Each side value must be {building_id: mesh_record}."""
if "cands" in self.od and "index" in self.od:
return ["cands", "index"], self.od
# Use whatever keys the dict already has as side names
return list(self.od.keys()), self.od
def _compute(self):
sides, od = self._resolve_sides()
result = {p: {s: {} for s in sides} for p in PROP_NAMES}
for side in sides:
for bid, rec in od[side].items():
faces, all_v = _extract_faces(rec)
vals = _compute_all(all_v, faces, rec)
for p in PROP_NAMES:
raw = vals.get(p, 0.0)
if self.vector_normalization and p not in ("num_vertices", "num_floors"):
raw = np.log1p(raw)
result[p][side][bid] = float(raw)
return result
|