File size: 30,943 Bytes
15d1f1d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 | #!/usr/bin/env python3
"""Robust semantic-view and primitive matching kernels for geometric v5.2.
The frozen v5/v5.1 evaluators estimate a sheet transform before they know
which primitives belong to which orthographic view. A missing view or a
changed circle can therefore perturb the shared scale and contaminate every
downstream component. This module reverses that order:
1. canonicalise and cluster each drawing independently;
2. match clusters to the audited GT semantic roles;
3. estimate one positive shared scale from the matched view envelopes;
4. compare geometry with a role-local translation and the shared scale.
No DXF handles or entity order are used for matching.
"""
from __future__ import annotations
import math
from itertools import combinations
from pathlib import Path
from typing import Any, Mapping, Sequence
import numpy as np
from scipy.optimize import linear_sum_assignment
try:
from . import eval_geometric05 as EV5
except ImportError: # direct evaluator import
import eval_geometric05 as EV5
SCHEMA_VERSION = "view-entity-matching-v5.2"
MAX_PREDICTED_VIEW_CANDIDATES = 8
ROLE_CENTER_ACCEPT_FRAC = 0.20
def _round(value: float | None, digits: int = 6) -> float | None:
if value is None or not math.isfinite(float(value)):
return None
return round(float(value), digits)
def _clip(value: float) -> float:
return float(min(1.0, max(0.0, value)))
def _empty_entities() -> dict[str, list[Any]]:
return {"lines": [], "circles": [], "arcs": [], "dims": []}
def _visibility(layer: Any) -> str:
token = str(layer).casefold()
return "hidden" if "hidden" in token or "dash" in token else "visible"
def _definition_rows(sidecar: Mapping[str, Any] | None) -> list[dict[str, Any]]:
rows = EV5._normalise_sidecar_views(sidecar)
names = [str(row.get("name")) for row in rows]
if len(rows) != EV5.VIEW_TARGET_COUNT or len(set(names)) != len(names):
return []
return rows
def _target_views_by_role(
canonical_gt: Mapping[str, Any], sidecar: Mapping[str, Any] | None
) -> tuple[dict[str, Mapping[str, Any]], str | None]:
definitions = _definition_rows(sidecar)
if not definitions:
return {}, "sidecar_reference_view_count_mismatch"
origin = np.zeros(2, dtype=float)
views = EV5._views_from_reference_boxes(
canonical_gt, definitions, origin, origin, 1.0
)
result: dict[str, Mapping[str, Any]] = {}
for view in views:
center = np.asarray(view["center"], dtype=float)
matches = []
for definition in definitions:
x0, y0, x1, y1 = map(float, definition["bbox"])
if (
min(x0, x1) - 1e-7 <= center[0] <= max(x0, x1) + 1e-7
and min(y0, y1) - 1e-7 <= center[1] <= max(y0, y1) + 1e-7
):
matches.append(str(definition["name"]))
if len(matches) != 1 or matches[0] in result:
return {}, "target_semantic_partition_ambiguous"
result[matches[0]] = view
expected = {str(row["name"]) for row in definitions}
if set(result) != expected:
return {}, "target_semantic_partition_incomplete"
return result, None
def _family_counts(entities: Mapping[str, Any]) -> dict[str, int]:
lines = list(entities.get("lines", []))
return {
"visible_line": sum(_visibility(line[-1]) == "visible" for line in lines),
"hidden_line": sum(_visibility(line[-1]) == "hidden" for line in lines),
"circle": len(entities.get("circles", [])),
"arc": len(entities.get("arcs", [])),
}
def _count_f1(predicted: int, target: int) -> float:
if predicted == 0 and target == 0:
return 1.0
if predicted == 0 or target == 0:
return 0.0
matched = min(predicted, target)
precision, recall = matched / predicted, matched / target
return float(2.0 * precision * recall / (precision + recall))
def _local_shape_score(
predicted: Mapping[str, Any], target: Mapping[str, Any]
) -> tuple[float, dict[str, Any]]:
pred_size = np.asarray(predicted["size"], dtype=float)
gt_size = np.asarray(target["size"], dtype=float)
pred_diag = float(np.linalg.norm(pred_size))
gt_diag = max(float(np.linalg.norm(gt_size)), 1e-9)
if pred_diag <= 1e-12:
return 0.0, {"reason": "degenerate_predicted_view"}
local_scale = gt_diag / pred_diag
strict = EV5.EV3.score_geometry(
predicted["entities"],
target["entities"],
np.asarray(predicted["center"], dtype=float),
np.asarray(target["center"], dtype=float),
local_scale,
max(gt_diag, 1.0),
)
dense = float(
EV5.EV3.dense_geometry_score(
predicted["entities"],
target["entities"],
np.asarray(predicted["center"], dtype=float),
np.asarray(target["center"], dtype=float),
local_scale,
max(gt_diag, 1.0),
)
)
pred_counts, gt_counts = _family_counts(predicted["entities"]), _family_counts(target["entities"])
active = [
_count_f1(pred_counts[name], gt_counts[name])
for name in pred_counts
if pred_counts[name] or gt_counts[name]
]
count_score = float(np.mean(active)) if active else 0.0
pred_nonzero = pred_size[pred_size > 1e-12]
gt_nonzero = gt_size[gt_size > 1e-12]
if len(pred_nonzero) == 2 and len(gt_nonzero) == 2:
pred_aspect = float(pred_nonzero[0] / pred_nonzero[1])
gt_aspect = float(gt_nonzero[0] / gt_nonzero[1])
aspect_score = math.exp(-abs(math.log(max(pred_aspect, 1e-12) / max(gt_aspect, 1e-12))))
else:
aspect_score = 1.0
score = _clip(
0.42 * dense
+ 0.33 * float(strict.get("score", 0.0))
+ 0.17 * count_score
+ 0.08 * aspect_score
)
return score, {
"score": _round(score),
"dense": _round(dense),
"strict": _round(float(strict.get("score", 0.0))),
"count": _round(count_score),
"aspect": _round(aspect_score),
"local_scale": _round(local_scale, 8),
}
def _positive_ratio(numerator: float, denominator: float) -> float | None:
if numerator <= 1e-12 or denominator <= 1e-12:
return None
value = float(numerator / denominator)
return value if math.isfinite(value) and value > 1e-8 else None
def _candidate_scales(
predicted: Sequence[Mapping[str, Any]], target: Sequence[Mapping[str, Any]]
) -> list[float]:
values: list[float] = []
for pred in predicted:
pred_size = np.asarray(pred["size"], dtype=float)
for gt in target:
gt_size = np.asarray(gt["size"], dtype=float)
for numerator, denominator in (
(float(np.linalg.norm(gt_size)), float(np.linalg.norm(pred_size))),
(float(gt_size[0]), float(pred_size[0])),
(float(gt_size[1]), float(pred_size[1])),
):
ratio = _positive_ratio(numerator, denominator)
if ratio is not None:
values.append(ratio)
pred_pairs = list(combinations(range(len(predicted)), 2))
target_pairs = list(combinations(range(len(target)), 2))
for first, second in pred_pairs:
pred_distance = float(
np.linalg.norm(
np.asarray(predicted[first]["center"], dtype=float)
- np.asarray(predicted[second]["center"], dtype=float)
)
)
for gt_first, gt_second in target_pairs:
gt_distance = float(
np.linalg.norm(
np.asarray(target[gt_first]["center"], dtype=float)
- np.asarray(target[gt_second]["center"], dtype=float)
)
)
ratio = _positive_ratio(gt_distance, pred_distance)
if ratio is not None:
values.append(ratio)
if not values:
return [1.0]
unique: dict[float, float] = {}
for value in values:
key = round(math.log(value), 6)
unique.setdefault(key, value)
return list(unique.values())
def _assignment(
predicted: Sequence[Mapping[str, Any]],
target: Sequence[Mapping[str, Any]],
shape: np.ndarray,
scale: float,
translation: np.ndarray,
sheet_diagonal: float,
) -> tuple[list[tuple[int, int]], np.ndarray, np.ndarray]:
pred_centers = np.stack([np.asarray(view["center"], dtype=float) for view in predicted])
gt_centers = np.stack([np.asarray(view["center"], dtype=float) for view in target])
transformed = pred_centers * float(scale) + np.asarray(translation, dtype=float)
center_error = np.linalg.norm(transformed[:, None, :] - gt_centers[None, :, :], axis=2) / max(sheet_diagonal, 1e-9)
cost = center_error + 0.16 * (1.0 - shape)
rows, columns = linear_sum_assignment(cost)
pairs = [(int(row), int(column)) for row, column in zip(rows, columns)]
return pairs, center_error, cost
def _fit_semantic_alignment(
predicted: Sequence[Mapping[str, Any]],
roles: Sequence[str],
target: Sequence[Mapping[str, Any]],
) -> dict[str, Any]:
if not predicted:
return {
"eligible": True,
"reason": "prediction_has_no_views",
"global_scale": 1.0,
"translation": np.zeros(2, dtype=float),
"role_to_pred": {},
"pairs": [],
"extra_predicted_indices": [],
"shape_details": {},
}
target_lows = np.stack([np.asarray(view["low"], dtype=float) for view in target])
target_highs = np.stack([np.asarray(view["high"], dtype=float) for view in target])
sheet_diagonal = max(
float(np.linalg.norm(np.max(target_highs, axis=0) - np.min(target_lows, axis=0))),
1.0,
)
shape = np.zeros((len(predicted), len(target)), dtype=float)
shape_details: dict[tuple[int, int], dict[str, Any]] = {}
for pred_index, pred_view in enumerate(predicted):
for gt_index, gt_view in enumerate(target):
value, detail = _local_shape_score(pred_view, gt_view)
shape[pred_index, gt_index] = value
shape_details[(pred_index, gt_index)] = detail
best: dict[str, Any] | None = None
for scale in _candidate_scales(predicted, target):
for pred_index, pred_view in enumerate(predicted):
pred_center = np.asarray(pred_view["center"], dtype=float)
for gt_index, gt_view in enumerate(target):
translation = np.asarray(gt_view["center"], dtype=float) - scale * pred_center
pairs: list[tuple[int, int]] = []
center_error = np.empty((len(predicted), len(target)), dtype=float)
cost = center_error
for _ in range(2):
pairs, center_error, cost = _assignment(
predicted, target, shape, scale, translation, sheet_diagonal
)
offsets = [
np.asarray(target[gt]["center"], dtype=float)
- scale * np.asarray(predicted[pred]["center"], dtype=float)
for pred, gt in pairs
]
if offsets:
translation = np.median(np.stack(offsets), axis=0)
pairs, center_error, cost = _assignment(
predicted, target, shape, scale, translation, sheet_diagonal
)
matched_center = [min(float(center_error[pred, gt]), 1.0) for pred, gt in pairs]
matched_shape = [float(shape[pred, gt]) for pred, gt in pairs]
missing = len(target) - len(pairs)
objective = (
0.72 * (sum(matched_center) + missing) / len(target)
+ 0.20 * (sum(1.0 - value for value in matched_shape) + missing) / len(target)
+ 0.08 * missing / len(target)
)
candidate = {
"objective": objective,
"scale": float(scale),
"translation": np.asarray(translation, dtype=float),
"pairs": pairs,
"center_error": center_error,
}
if best is None or candidate["objective"] < best["objective"]:
best = candidate
assert best is not None
# Re-estimate the shared scale from matched view envelopes and inter-view
# distances. Taking the median in log space prevents one mutated circle,
# arc, or locally scaled view from setting the scale for the whole sheet.
ratios: list[float] = []
for pred_index, gt_index in best["pairs"]:
pred_size = np.asarray(predicted[pred_index]["size"], dtype=float)
gt_size = np.asarray(target[gt_index]["size"], dtype=float)
for axis in range(2):
ratio = _positive_ratio(float(gt_size[axis]), float(pred_size[axis]))
if ratio is not None:
ratios.append(ratio)
for (pred_a, gt_a), (pred_b, gt_b) in combinations(best["pairs"], 2):
pred_distance = float(
np.linalg.norm(
np.asarray(predicted[pred_a]["center"], dtype=float)
- np.asarray(predicted[pred_b]["center"], dtype=float)
)
)
gt_distance = float(
np.linalg.norm(
np.asarray(target[gt_a]["center"], dtype=float)
- np.asarray(target[gt_b]["center"], dtype=float)
)
)
ratio = _positive_ratio(gt_distance, pred_distance)
if ratio is not None:
ratios.append(ratio)
scale = (
float(math.exp(float(np.median(np.log(np.asarray(ratios, dtype=float))))))
if ratios
else float(best["scale"])
)
offsets = [
np.asarray(target[gt]["center"], dtype=float)
- scale * np.asarray(predicted[pred]["center"], dtype=float)
for pred, gt in best["pairs"]
]
translation = np.median(np.stack(offsets), axis=0) if offsets else np.zeros(2, dtype=float)
pairs, center_error, _ = _assignment(
predicted, target, shape, scale, translation, sheet_diagonal
)
accepted = [
(pred, gt)
for pred, gt in pairs
if float(center_error[pred, gt]) <= ROLE_CENTER_ACCEPT_FRAC
]
role_to_pred = {roles[gt]: pred for pred, gt in accepted}
used = {pred for pred, _ in accepted}
pair_rows = [
{
"role": roles[gt],
"predicted_index": pred,
"target_index": gt,
"accepted": (pred, gt) in accepted,
"center_error_fraction": _round(float(center_error[pred, gt])),
"shape_score": _round(float(shape[pred, gt])),
"shape": shape_details[(pred, gt)],
}
for pred, gt in pairs
]
return {
"eligible": True,
"reason": "ok" if len(accepted) == len(target) else "semantic_roles_missing",
"global_scale": scale,
"translation": translation,
"sheet_diagonal": sheet_diagonal,
"role_to_pred": role_to_pred,
"pairs": pair_rows,
"extra_predicted_indices": sorted(set(range(len(predicted))) - used),
"shape_details": shape_details,
"objective": float(best["objective"]),
}
def build_pair_context(
pred: str | Path,
gt: str | Path,
*,
sidecar: Mapping[str, Any] | None,
) -> dict[str, Any]:
pred_path, gt_path = Path(pred), Path(gt)
pred_meta, gt_meta = EV5._read_structure(pred_path), EV5._read_structure(gt_path)
raw_pred = EV5._extract_entities(pred_path) if pred_meta.get("readable") else None
raw_gt = EV5._extract_entities(gt_path) if gt_meta.get("readable") else None
target_valid = bool(raw_gt and EV5._n_geometry(raw_gt) and EV5._safe_bbox(raw_gt) is not None)
prediction_valid = bool(raw_pred and EV5._n_geometry(raw_pred) and EV5._safe_bbox(raw_pred) is not None)
context: dict[str, Any] = {
"schema_version": SCHEMA_VERSION,
"target_valid": target_valid,
"prediction_valid": prediction_valid,
"pred_meta": pred_meta,
"gt_meta": gt_meta,
"raw_pred": raw_pred,
"raw_gt": raw_gt,
}
if not target_valid:
context.update({"eligible": False, "reason": "target_dxf_unreadable_or_degenerate"})
return context
canonical_gt = EV5.EV4.canonicalize_entities(raw_gt)
gt_by_role, error = _target_views_by_role(canonical_gt, sidecar)
if error:
context.update(
{
"eligible": False,
"reason": error,
"canonical_gt": canonical_gt,
}
)
return context
context["canonical_gt"] = canonical_gt
context["gt_by_role"] = gt_by_role
if not prediction_valid:
context.update(
{
"eligible": True,
"reason": "prediction_dxf_unreadable_or_degenerate",
"canonical_pred": _empty_entities(),
"pred_views": [],
"alignment": _fit_semantic_alignment([], list(gt_by_role), list(gt_by_role.values())),
}
)
return context
canonical_pred = EV5.EV4.canonicalize_entities(raw_pred)
pred_views_all = EV5._cluster_views(canonical_pred)
ranked = sorted(
enumerate(pred_views_all),
key=lambda item: (
int(item[1].get("n_primitives", 0)),
float(np.linalg.norm(np.asarray(item[1].get("size", (0.0, 0.0)), dtype=float))),
),
reverse=True,
)
selected_pairs = ranked[:MAX_PREDICTED_VIEW_CANDIDATES]
pred_views = [view for _, view in selected_pairs]
selected_original_indices = [index for index, _ in selected_pairs]
roles = list(gt_by_role)
alignment = _fit_semantic_alignment(pred_views, roles, [gt_by_role[role] for role in roles])
alignment["selected_original_indices"] = selected_original_indices
alignment["omitted_original_indices"] = [
index for index in range(len(pred_views_all)) if index not in selected_original_indices
]
context.update(
{
"eligible": True,
"reason": alignment["reason"],
"canonical_pred": canonical_pred,
"pred_views": pred_views,
"pred_views_all": pred_views_all,
"alignment": alignment,
}
)
return context
def role_local_transform(
context: Mapping[str, Any], role: str
) -> tuple[float, np.ndarray] | None:
alignment = context.get("alignment", {})
role_to_pred = alignment.get("role_to_pred", {}) if isinstance(alignment, Mapping) else {}
pred_index = role_to_pred.get(role) if isinstance(role_to_pred, Mapping) else None
gt_by_role = context.get("gt_by_role", {})
pred_views = context.get("pred_views", [])
if not isinstance(pred_index, int) or role not in gt_by_role or not 0 <= pred_index < len(pred_views):
return None
scale = float(alignment.get("global_scale", 1.0))
translation = (
np.asarray(gt_by_role[role]["center"], dtype=float)
- scale * np.asarray(pred_views[pred_index]["center"], dtype=float)
)
return scale, translation
def _prf(matched: int, predicted: int, target: int) -> dict[str, Any]:
eligible = bool(predicted or target)
if not eligible:
return {
"eligible": False,
"matched": 0,
"predicted": 0,
"target": 0,
"precision": None,
"recall": None,
"f1": None,
}
precision = matched / predicted if predicted else 0.0
recall = matched / target if target else 1.0
f1 = 0.0 if precision + recall <= 0.0 else 2.0 * precision * recall / (precision + recall)
return {
"eligible": True,
"matched": int(matched),
"predicted": int(predicted),
"target": int(target),
"precision": _round(precision),
"recall": _round(recall),
"f1": _round(f1),
}
def _line_array(line: Sequence[Any], scale: float, translation: np.ndarray) -> np.ndarray:
points = np.asarray([[line[0], line[1]], [line[2], line[3]]], dtype=float)
return points * scale + translation
def _line_error(first: np.ndarray, second: np.ndarray, diagonal: float) -> float:
direct = float(np.linalg.norm(first[0] - second[0]) + np.linalg.norm(first[1] - second[1]))
reverse = float(np.linalg.norm(first[0] - second[1]) + np.linalg.norm(first[1] - second[0]))
return min(direct, reverse) / (2.0 * max(diagonal, 1e-9))
def _match_lines(
predicted: Sequence[Sequence[Any]],
target: Sequence[Sequence[Any]],
*,
scale: float,
translation: np.ndarray,
diagonal: float,
) -> dict[str, Any]:
if not predicted or not target:
result = _prf(0, len(predicted), len(target))
result.update({"mean_endpoint_error": None, "worst_endpoint_error": None, "pairs": []})
return result
pred_arrays = [_line_array(line, scale, translation) for line in predicted]
gt_arrays = [_line_array(line, 1.0, np.zeros(2, dtype=float)) for line in target]
costs = np.asarray(
[[_line_error(pred, gt, diagonal) for gt in gt_arrays] for pred in pred_arrays],
dtype=float,
)
rows, columns = linear_sum_assignment(costs)
pairs = []
errors = []
for row, column in zip(rows, columns):
error = float(costs[row, column])
if error <= 0.02:
errors.append(error)
pairs.append(
{
"predicted_index": int(row),
"target_index": int(column),
"endpoint_error_fraction": _round(error),
}
)
result = _prf(len(pairs), len(predicted), len(target))
result.update(
{
"mean_endpoint_error": _round(float(np.mean(errors))) if errors else None,
"worst_endpoint_error": _round(max(errors)) if errors else None,
"parameter_score": _round(math.exp(-float(np.mean(errors)) / 0.01)) if errors else (0.0 if target else None),
"pairs": pairs,
}
)
return result
def _circle_record(circle: Sequence[Any], scale: float, translation: np.ndarray) -> tuple[np.ndarray, float, str]:
center = np.asarray([circle[0], circle[1]], dtype=float) * scale + translation
return center, abs(float(circle[2])) * scale, _visibility(circle[-1])
def _match_circles(
predicted: Sequence[Sequence[Any]],
target: Sequence[Sequence[Any]],
*,
scale: float,
translation: np.ndarray,
diagonal: float,
) -> dict[str, Any]:
if not predicted or not target:
result = _prf(0, len(predicted), len(target))
result.update({"mean_radius_log_error": None, "worst_radius_log_error": None, "parameter_score": 0.0 if target else None, "pairs": []})
return result
pred_records = [_circle_record(item, scale, translation) for item in predicted]
gt_records = [_circle_record(item, 1.0, np.zeros(2, dtype=float)) for item in target]
costs = np.zeros((len(predicted), len(target)), dtype=float)
raw: dict[tuple[int, int], tuple[float, float, float, bool]] = {}
for p, (pred_center, pred_radius, pred_visibility) in enumerate(pred_records):
for g, (gt_center, gt_radius, gt_visibility) in enumerate(gt_records):
center_error = float(np.linalg.norm(pred_center - gt_center) / max(diagonal, 1e-9))
log_error = abs(math.log(max(pred_radius, 1e-12) / max(gt_radius, 1e-12)))
relative_error = abs(pred_radius / max(gt_radius, 1e-12) - 1.0)
visibility_ok = pred_visibility == gt_visibility
costs[p, g] = center_error + 0.35 * log_error + (0.5 if not visibility_ok else 0.0)
raw[(p, g)] = (center_error, log_error, relative_error, visibility_ok)
rows, columns = linear_sum_assignment(costs)
pairs, log_errors = [], []
for row, column in zip(rows, columns):
center_error, log_error, relative_error, visibility_ok = raw[(int(row), int(column))]
if center_error <= 0.08 and log_error <= math.log(2.0) and visibility_ok:
log_errors.append(log_error)
pairs.append(
{
"predicted_index": int(row),
"target_index": int(column),
"center_error_fraction": _round(center_error),
"radius_log_error": _round(log_error),
"radius_relative_error": _round(relative_error),
"diameter_relative_error": _round(relative_error),
}
)
result = _prf(len(pairs), len(predicted), len(target))
result.update(
{
"mean_radius_log_error": _round(float(np.mean(log_errors))) if log_errors else None,
"worst_radius_log_error": _round(max(log_errors)) if log_errors else None,
"parameter_score": _round(math.exp(-float(np.mean(log_errors)) / 0.10)) if log_errors else (0.0 if target else None),
"pairs": pairs,
}
)
return result
def _angle_distance(first: float, second: float) -> float:
return abs((float(first) - float(second) + 180.0) % 360.0 - 180.0)
def _arc_record(arc: Sequence[Any], scale: float, translation: np.ndarray) -> tuple[np.ndarray, float, float, float, float, str]:
center = np.asarray([arc[0], arc[1]], dtype=float) * scale + translation
start, end = float(arc[3]) % 360.0, float(arc[4]) % 360.0
sweep = (end - start) % 360.0
if sweep <= 1e-9:
sweep = 360.0
return center, abs(float(arc[2])) * scale, start, end, sweep, _visibility(arc[-1])
def _match_arcs(
predicted: Sequence[Sequence[Any]],
target: Sequence[Sequence[Any]],
*,
scale: float,
translation: np.ndarray,
diagonal: float,
) -> dict[str, Any]:
if not predicted or not target:
result = _prf(0, len(predicted), len(target))
result.update({"mean_radius_log_error": None, "mean_sweep_error_fraction": None, "parameter_score": 0.0 if target else None, "pairs": []})
return result
pred_records = [_arc_record(item, scale, translation) for item in predicted]
gt_records = [_arc_record(item, 1.0, np.zeros(2, dtype=float)) for item in target]
costs = np.zeros((len(predicted), len(target)), dtype=float)
raw: dict[tuple[int, int], tuple[float, float, float, float, bool]] = {}
for p, pred in enumerate(pred_records):
for g, target_item in enumerate(gt_records):
center_error = float(np.linalg.norm(pred[0] - target_item[0]) / max(diagonal, 1e-9))
radius_error = abs(math.log(max(pred[1], 1e-12) / max(target_item[1], 1e-12)))
radius_relative = abs(pred[1] / max(target_item[1], 1e-12) - 1.0)
sweep_error = abs(pred[4] - target_item[4]) / 360.0
angle_error = (_angle_distance(pred[2], target_item[2]) + _angle_distance(pred[3], target_item[3])) / 720.0
visibility_ok = pred[5] == target_item[5]
costs[p, g] = center_error + 0.30 * radius_error + 0.25 * sweep_error + 0.15 * angle_error + (0.5 if not visibility_ok else 0.0)
raw[(p, g)] = (center_error, radius_error, radius_relative, sweep_error, visibility_ok)
rows, columns = linear_sum_assignment(costs)
pairs, parameter_errors = [], []
for row, column in zip(rows, columns):
center_error, radius_error, radius_relative, sweep_error, visibility_ok = raw[(int(row), int(column))]
if center_error <= 0.08 and radius_error <= math.log(2.0) and sweep_error <= 0.25 and visibility_ok:
combined = radius_error + sweep_error
parameter_errors.append(combined)
pairs.append(
{
"predicted_index": int(row),
"target_index": int(column),
"center_error_fraction": _round(center_error),
"radius_log_error": _round(radius_error),
"radius_relative_error": _round(radius_relative),
"sweep_error_fraction": _round(sweep_error),
}
)
result = _prf(len(pairs), len(predicted), len(target))
result.update(
{
"mean_radius_log_error": _round(float(np.mean([row["radius_log_error"] for row in pairs]))) if pairs else None,
"mean_sweep_error_fraction": _round(float(np.mean([row["sweep_error_fraction"] for row in pairs]))) if pairs else None,
"parameter_score": _round(math.exp(-float(np.mean(parameter_errors)) / 0.10)) if parameter_errors else (0.0 if target else None),
"pairs": pairs,
}
)
return result
def match_view_primitives(
predicted_view: Mapping[str, Any] | None,
target_view: Mapping[str, Any],
*,
global_scale: float,
) -> dict[str, Any]:
target_entities = target_view["entities"]
predicted_entities = predicted_view["entities"] if predicted_view is not None else _empty_entities()
diagonal = max(float(np.linalg.norm(np.asarray(target_view["size"], dtype=float))), 1.0)
if predicted_view is None:
translation = np.zeros(2, dtype=float)
else:
translation = (
np.asarray(target_view["center"], dtype=float)
- global_scale * np.asarray(predicted_view["center"], dtype=float)
)
pred_visible = [line for line in predicted_entities.get("lines", []) if _visibility(line[-1]) == "visible"]
gt_visible = [line for line in target_entities.get("lines", []) if _visibility(line[-1]) == "visible"]
pred_hidden = [line for line in predicted_entities.get("lines", []) if _visibility(line[-1]) == "hidden"]
gt_hidden = [line for line in target_entities.get("lines", []) if _visibility(line[-1]) == "hidden"]
return {
"visible_line": _match_lines(
pred_visible, gt_visible, scale=global_scale, translation=translation, diagonal=diagonal
),
"hidden_line": _match_lines(
pred_hidden, gt_hidden, scale=global_scale, translation=translation, diagonal=diagonal
),
"circle": _match_circles(
list(predicted_entities.get("circles", [])),
list(target_entities.get("circles", [])),
scale=global_scale,
translation=translation,
diagonal=diagonal,
),
"arc": _match_arcs(
list(predicted_entities.get("arcs", [])),
list(target_entities.get("arcs", [])),
scale=global_scale,
translation=translation,
diagonal=diagonal,
),
"local_translation": [float(value) for value in translation],
"target_diagonal": diagonal,
}
__all__ = [
"MAX_PREDICTED_VIEW_CANDIDATES",
"ROLE_CENTER_ACCEPT_FRAC",
"SCHEMA_VERSION",
"build_pair_context",
"match_view_primitives",
"role_local_transform",
]
|