Spaces:
Running on Zero
Running on Zero
File size: 7,244 Bytes
2874635 | 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 | #!/usr/bin/env python3
from __future__ import annotations
from typing import TYPE_CHECKING
import numpy as np
if TYPE_CHECKING:
from spatial_logic import Detection
CARTOUCHE_CLASS = 'cartouche'
EDGE_FRAC = 0.02 # touching this fraction of the image border -> drop
# the cartouche from the aspect vote (clipped)
MIN_SIGNAL_GAP = 0.10 # column-vs-row vote margin below which we tie
def _cluster_1d_bands(values: np.ndarray, bandwidth: float) -> int:
"""1-D running-mean clustering. Returns the number of clusters.
Mirrors spatial_logic._group_lines so the band count matches what
the reading-order assembler would see.
"""
if len(values) == 0:
return 0
v = np.sort(values)
bands = [[v[0]]]
for x in v[1:]:
mean_b = float(np.mean(bands[-1]))
if x - mean_b > bandwidth:
bands.append([x])
else:
bands[-1].append(x)
return len(bands)
def _vote_cartouche_aspect(
detections : list[Detection],
img_w: float,
img_h: float,
) -> tuple[str | None, float]:
"""Tall cartouches -> columns; wide -> rows.
Cartouches whose bbox touches the image border are dropped (clipped)."""
edge_x = EDGE_FRAC * img_w
edge_y = EDGE_FRAC * img_h
tall = wide = 0
for d in detections:
if not getattr(d, 'is_cartouche', lambda: False)():
continue
x1, y1, x2, y2 = d.bbox
if x1 <= edge_x or y1 <= edge_y or x2 >= img_w - edge_x or y2 >= img_h - edge_y:
continue
w, h = x2 - x1, y2 - y1
if h > 1.2 * w:
tall += 1
elif w > 1.2 * h:
wide += 1
total = tall + wide
if total == 0:
return None, 0.0
if tall > wide:
# tie -> historical default
return 'columns', tall / total
if wide > tall:
# tie -> historical default
return 'rows', wide / total
return None, 0.0
def _vote_band_count(
detections : list[Detection],
img_w: float,
img_h: float,
) -> tuple[str | None, float]:
"""Cluster non-cartouche centroids on x and on y; reading axis is
the one with MORE bands (signs stacked along it)."""
sign_dets = [d for d in detections
if not getattr(d, 'is_cartouche', lambda: False)()]
if len(sign_dets) < 4:
return None, 0.0
cx = np.array([d.centroid[0] for d in sign_dets])
cy = np.array([d.centroid[1] for d in sign_dets])
med_w = float(np.median([d.width for d in sign_dets]))
med_h = float(np.median([d.height for d in sign_dets]))
# bands_x : n_x: how many distinct vertical bands of signs (potential columns)
# bands_y : n_y: how many distinct horizontal bands of signs (potential rows)
n_x = _cluster_1d_bands(cx, 0.6 * med_w) # columns count
n_y = _cluster_1d_bands(cy, 0.6 * med_h) # rows count
if n_x == 0 or n_y == 0:
return None, 0.0
ratio = abs(n_y - n_x) / max(n_x, n_y)
if n_y > n_x: # more rows than columns -> text in columns
return 'columns', ratio
if n_x > n_y:
return 'rows', ratio
return None, 0.0
def _vote_sign_aspect(
detections : list[Detection],
img_w: float,
img_h: float
) -> tuple[str | None, float]:
"""Median sign aspect ratio. Mild tiebreaker."""
aspects = []
for d in detections:
if getattr(d, 'is_cartouche', lambda: False)():
continue
if d.width <= 0 or d.height <= 0:
continue
aspects.append(d.height / d.width)
if not aspects:
return None, 0.0
med = float(np.median(aspects))
if med >= 1.15:
return 'columns', min(0.3, (med - 1.0))
if med <= 0.85:
return 'rows', min(0.3, (1.0 - med))
return None, 0.0
def detect_layout_from_detections(
detections : list[Detection],
img_w: float,
img_h: float, *,
verbose: bool = False,
) -> str:
"""
Vote across three geometric signals and return 'rows' or 'columns'.
Fragile-breakpoint fallback: 'rows' on no evidence (historical default).
"""
if not detections:
if verbose:
print(f"WARNING: no detections")
# fallback to historical default rows layout
return 'rows'
if img_w <= 0 or img_h <= 0:
if verbose:
print(f"WARNING: invalid image size: {img_w}x{img_h}")
# fallback to historical default rows layout
raise ValueError(f"invalid image size: {img_w}x{img_h}")
return 'rows'
votes: dict[str, float] = {'rows': 0.0, 'columns': 0.0}
# More important signals have higher weights
weights = {'cartouche_aspect': 1.5, 'band_count': 1.0, 'sign_aspect': 0.5}
for name, fn in (
('cartouche_aspect', _vote_cartouche_aspect),
('band_count', _vote_band_count),
('sign_aspect', _vote_sign_aspect),
):
if name == 'sign_aspect':
label, conf = fn(detections, img_w, img_h)
else:
label, conf = fn(detections, img_w, img_h)
if label is not None:
votes[label] += weights[name] * conf
if verbose:
print(f" layout vote [{name}]: {label} (conf={conf:.2f})")
diff = votes['columns'] - votes['rows']
if verbose:
print(f" totals: columns={votes['columns']:.2f} rows={votes['rows']:.2f}")
if abs(diff) < MIN_SIGNAL_GAP:
return 'rows' # tie -> historical default
return 'columns' if diff > 0 else 'rows'
# Legacy pixel fallback (kept for callers that don't have detections yet).
# def detect_layout(img) -> str:
# """Pixel-based fallback (morphological line detection).
# Prefer detect_layout_from_detections() — this one fails on stone reliefs.
# """
# import cv2
# gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# binary = cv2.adaptiveThreshold(
# gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
# cv2.THRESH_BINARY_INV, 15, 4,
# )
# h, w = binary.shape
# h_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (w // 3, 1))
# v_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, h // 3))
# h_score = np.sum(cv2.morphologyEx(binary, cv2.MORPH_OPEN, h_kernel)) / 255
# v_score = np.sum(cv2.morphologyEx(binary, cv2.MORPH_OPEN, v_kernel)) / 255
# MIN_SCORE = 500
# if h_score > MIN_SCORE and h_score > v_score * 1.5:
# return 'rows'
# if v_score > MIN_SCORE and v_score > h_score * 1.5:
# return 'columns'
# return 'rows'
|