Image Segmentation
PyTorch
English
painting-vision-robotics-kit
semantic-segmentation
robotics
edge-ai
construction-ai
autonomous-painting
wall-painting-robot
paint-coverage-estimation
building-facade
drywall
skirting-detection
window-detection
lidar
depth-validation
segformer
deeplabv3
Eval Results (legacy)
File size: 118,080 Bytes
53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 4eecd03 53f4112 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 | #!/usr/bin/env python3
"""Operator console: photo in, reviewed and approved paint work order out.
A local web app for the person standing next to the robot. Upload (or take) a photo
of the wall; the kit shows where it will paint, what it keeps out of, which damage
needs prep first, and why the safety gate would stop. The operator can:
* toggle layers (paint area, keep-out, uncertain wall, touch-up, damage, tool paths);
* correct the plan with a brush: **paint here** / **do not paint**;
* sign the confirmations the gate allows a person to make (not wallpaper, depth
waived, uncertain areas reviewed), each recorded with their name;
* re-plan, then **approve**. An approved order is frozen: the job keeps the photo,
masks, edits and work order with their sha256, and an entry is appended to the
hash-linked ``registry.jsonl`` (``run_registry.py verify`` checks it).
Production notes: binds to 127.0.0.1 unless ``--host`` says otherwise; one model
loaded once, inference serialised behind a lock; uploads capped (``MAX_UPLOAD``)
and decoded before anything is written; job ids are random and file serving is an
allowlist, so no path from a request reaches the filesystem. stdlib + the kit's own
dependencies only: nothing extra to install on the robot PC.
Usage::
python3 console_server.py --checkpoint artifacts/best.pt --jobs-dir jobs
# open http://127.0.0.1:8765
"""
from __future__ import annotations
import argparse
import base64
import hashlib
import io
import json
import re
import secrets
import threading
import time
from http import HTTPStatus
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from pathlib import Path
from urllib.parse import parse_qs, urlparse
import numpy as np
from PIL import Image
from work_order import CONFIRMATIONS, EDIT_ADD, EDIT_EXCLUDE, apply_edits, build_work_order, scale_from_prediction
VERSION = "1.0"
STATIC = Path(__file__).with_name("console")
MAX_UPLOAD = 30 * 1024 * 1024
MAX_SIDE = 2048 # frames are normalised to this long side before analysis
JOB_ID = re.compile(r"^[0-9a-f]{12}$")
LAYER_COLOURS = {"paint": (46, 204, 113), "keepout": (231, 76, 60), "uncertain": (241, 196, 15),
"touchup": (0, 210, 255)}
JOB_FILES = {"photo.jpg": "image/jpeg", "demo_paint.png": "image/png", "demo_regions.png": "image/png", "work_order.json": "application/json", "result.json": "application/json",
"approval.json": "application/json", "layer_edits.png": "image/png",
**{f"layer_{k}.png": "image/png" for k in LAYER_COLOURS},
**{f"layer_{k}_edge.png": "image/png" for k in LAYER_COLOURS},
"layer_floor.png": "image/png", "layer_floor_edge.png": "image/png", "floor.json": "application/json",
"dronetwin_heatmap.png": "image/png", "dronetwin.json": "application/json", "dronetwin_geojson.json": "application/geo+json"}
FLOOR_COLOUR = (176, 125, 79)
VIDEO_ID = re.compile(r"^v[0-9a-f]{12}$")
VIDEO_EXT = {"mp4": "video/mp4", "mov": "video/quicktime", "webm": "video/webm", "m4v": "video/mp4"}
VIDEO_FILE = re.compile(r"^(manifest\.json|track\.json|plan_\d+\.json|frame_\d+\.png|paint_\d+\.png|photo_\d+\.jpg|video\.(mp4|mov|webm|m4v))$")
MAX_VIDEO = 200 * 1024 * 1024
MAX_DRAWING = 40 * 1024 * 1024
MAX_DEPTH = 60 * 1024 * 1024
MAX_SCAN = 150 * 1024 * 1024
MAX_VIEWS = 14
ZONE_KINDS = ("keepout", "paint", "note", "window", "door", "fixture", "wall")
CORRECT_CLASS = {"window": 8, "door": 7, "fixture": 9} # operator corrections: "this is a ..." (fixture = wall obstacle)
CORRECTION_KINDS = ("window", "door", "fixture", "wall")
ZONE_SHAPES = ("rect", "poly", "pin")
MAX_ZONES = 200
INFO_FIELDS = ("client", "site", "room", "spec", "notes")
DRAWING_ID = re.compile(r"^d[0-9a-f]{8}$")
DRAWING_SIDE = 2400 # a drawing is kept at this long side: enough to click a dimension on
STATIC_FILES = {"index.html": "text/html; charset=utf-8", "app.js": "text/javascript; charset=utf-8",
"style.css": "text/css; charset=utf-8", "demo.html": "text/html; charset=utf-8",
"demo.js": "text/javascript; charset=utf-8", "demo.css": "text/css; charset=utf-8",
"demo_video.js": "text/javascript; charset=utf-8", "zones.js": "text/javascript; charset=utf-8",
"zones.css": "text/css; charset=utf-8", "floor.js": "text/javascript; charset=utf-8", "dronetwin.js": "text/javascript; charset=utf-8", "robot.js": "text/javascript; charset=utf-8", "collapse.js": "text/javascript; charset=utf-8", "collapse.css": "text/css; charset=utf-8", "live.js": "text/javascript; charset=utf-8", "twin.js": "text/javascript; charset=utf-8", "prompt.js": "text/javascript; charset=utf-8"}
def _points(pts):
"""Two clicked [x, y] points, or None: stored so the drawn lines come back when a drawing is reopened."""
try:
(a, b), (c, d) = pts
return [[float(a), float(b)], [float(c), float(d)]]
except (TypeError, ValueError):
return None
def sha256_bytes(data):
return hashlib.sha256(data).hexdigest()
def now():
return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
def save_mask(path, mask):
Image.fromarray(np.asarray(mask, bool).astype(np.uint8) * 255).save(path)
def load_mask(path):
return np.asarray(Image.open(path).convert("L")) > 127
def layer_png(mask, colour, alpha=150):
mask = np.asarray(mask, bool)
rgba = np.zeros(mask.shape + (4,), np.uint8)
rgba[mask] = (*colour, alpha)
return Image.fromarray(rgba)
def edge_png(mask, colour, width=3):
"""The outline of a layer, legible on ANY wall colour: a ``width`` px ring in the layer's colour, a 2 px white halo on both sides of it, and a
thin dark edge outside the halo (the way a map outlines a road). A green fill vanishes on a green wall; this ring does not. Tiny specks are dropped."""
from scipy import ndimage
mask = np.asarray(mask, bool)
if mask.any():
mask = ndimage.binary_opening(mask, iterations=1)
inner = ndimage.binary_erosion(mask, iterations=width) if mask.any() else mask
ring = mask & ~inner
halo = ndimage.binary_dilation(ring, iterations=2) & ~ring
dark = ndimage.binary_dilation(ring, iterations=3) & ~ndimage.binary_dilation(ring, iterations=2)
rgba = np.zeros(mask.shape + (4,), np.uint8)
rgba[dark] = (24, 26, 32, 175)
rgba[halo] = (255, 255, 255, 235)
rgba[ring] = (*colour, 255)
return Image.fromarray(rgba)
def decode_edits(data_url, shape):
"""Edit canvas (PNG data URL) -> 0/1/2 layer: green = paint here, red = do not paint."""
if not data_url:
return None
match = re.match(r"^data:image/png;base64,([A-Za-z0-9+/=]+)$", data_url)
if not match:
raise ValueError("edits must be a PNG data URL")
image = Image.open(io.BytesIO(base64.b64decode(match.group(1)))).convert("RGBA")
if image.size != (shape[1], shape[0]):
image = image.resize((shape[1], shape[0]), Image.Resampling.NEAREST)
rgba = np.asarray(image)
painted = rgba[..., 3] > 0
edits = np.zeros(shape, np.uint8)
edits[painted & (rgba[..., 1] > rgba[..., 0])] = EDIT_ADD
edits[painted & (rgba[..., 0] >= rgba[..., 1])] = EDIT_EXCLUDE
return edits
def window_gate(rgb, classes):
"""Turn window blobs that are pictures, posters, mirrors, screens or boards (or have no frame) into uncertain wall:
window_negatives.py. Fails open to the ungated map, with the reason, if the general model cannot load."""
try:
import window_negatives as wn
gated, reasons = wn.filter_windows(np.asarray(classes), rgb, wn.ade_maps(rgb))
return gated, {"applied": True, "rejected": reasons}
except Exception as exc: # no transformers / no network for the first download
return classes, {"applied": False, "reason": str(exc)[:200]}
class Console:
"""Jobs on disk + the analyser. ``analyze_fn(rgb) -> (summary, masks)`` is injected (tests fake it)."""
def __init__(self, jobs_dir, analyze_fn, model_info, defaults=None):
self.jobs = Path(jobs_dir)
self.jobs.mkdir(parents=True, exist_ok=True)
self.analyze_fn = analyze_fn
self.model_info = model_info
self.defaults = defaults or {}
self.lock = threading.Lock() # one inference at a time; also serialises job writes
# ---------------------------------------------------------------- jobs
def job_dir(self, job_id):
if not JOB_ID.match(job_id or ""):
raise KeyError(job_id)
path = self.jobs / job_id
if not path.is_dir():
raise KeyError(job_id)
return path
def read_job(self, job_id):
path = self.job_dir(job_id)
job = json.loads((path / "job.json").read_text())
order_path = path / "work_order.json"
job["work_order"] = json.loads(order_path.read_text()) if order_path.is_file() else None
result = json.loads((path / "result.json").read_text())
job["regions"] = (result.get("surface_condition") or {}).get("regions", [])
job["measurements"] = result.get("measurements")
job["completion"] = result.get("completion")
job["paint_color"] = result.get("paint_color")
job["site_objects"] = result.get("site_objects")
job["has_edits"] = (path / "layer_edits.png").is_file()
return job
# ---------------------------------------------------------------- operator zones, notes and job information
def _clean_zones(self, path, request):
"""Validate what the console sent: zones are polygons / rectangles / pins in photo pixels."""
job = json.loads((path / "job.json").read_text())
w, h = float(job["width"]), float(job["height"])
zones = []
for z in (request.get("zones") or [])[:MAX_ZONES]:
kind, shape = z.get("kind"), z.get("shape")
if kind not in ZONE_KINDS or shape not in ZONE_SHAPES:
raise ValueError(f"unknown zone kind/shape: {kind!r}/{shape!r}")
pts = [[min(max(float(x), 0.0), w), min(max(float(y), 0.0), h)] for x, y in z.get("pts") or []]
need = {"pin": 1, "rect": 2, "poly": 3}[shape]
if len(pts) < need or len(pts) > 400:
raise ValueError(f"a {shape} zone needs {need}{'+' if shape == 'poly' else ''} points")
if shape == "pin" and kind != "note":
raise ValueError("a pin can only be a note")
zones.append({"id": str(z.get("id") or secrets.token_hex(3))[:12], "kind": kind, "shape": shape, "pts": pts[:2] if shape == "rect" else pts,
"label": str(z.get("label") or "")[:60], "text": str(z.get("text") or "")[:600],
"visible": bool(z.get("visible", True))})
info = {k: str((request.get("info") or {}).get(k) or "")[:600] for k in INFO_FIELDS}
return zones, info
def get_zones(self, job_id):
path = self.job_dir(job_id)
f = path / "zones.json"
return json.loads(f.read_text()) if f.is_file() else {"zones": [], "info": {k: "" for k in INFO_FIELDS}}
def set_zones(self, job_id, request):
path = self.job_dir(job_id)
if json.loads((path / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to change zones")
zones, info = self._clean_zones(path, request)
data = {"zones": zones, "info": info, "updated": now()}
(path / "zones.json").write_text(json.dumps(data, indent=1))
return data
@staticmethod
def zone_edits(zones, shape):
"""Rasterise keep-out / paint-only zones into an edits map (the same codes as the brush); notes do nothing."""
import cv2
out = np.zeros(shape, np.uint8)
# "plain wall" corrections free the area to paint; windows / doors / fixtures keep paint out; keep-out wins last
for kinds, code in ((("paint", "wall"), EDIT_ADD), (("keepout", "window", "door", "fixture"), EDIT_EXCLUDE)):
layer = np.zeros(shape, np.uint8)
for z in zones:
if z["kind"] not in kinds or z["shape"] == "pin":
continue
pts = np.array(z["pts"], np.float32)
if z["shape"] == "rect":
(x0, y0), (x1, y1) = pts
pts = np.array([[x0, y0], [x1, y0], [x1, y1], [x0, y1]], np.float32)
cv2.fillPoly(layer, [np.round(pts).astype(np.int32)], 1)
out[layer > 0] = code
return out
@staticmethod
def corrected_classes(classes, zones):
"""The model's class map with the operator's corrections painted in: "plain wall" zones erase detections,
window / door / fixture zones add them. Same ids as the segmenter (demo_plan.CLASS_IDS)."""
import cv2
out = np.asarray(classes).copy()
for kinds in (("wall",), tuple(CORRECT_CLASS)):
for z in zones:
if z["kind"] not in kinds or z["shape"] == "pin" or z.get("visible", True) is False:
continue
pts = np.array(z["pts"], np.float32)
if z["shape"] == "rect":
(x0, y0), (x1, y1) = pts
pts = np.array([[x0, y0], [x1, y0], [x1, y1], [x0, y1]], np.float32)
m = np.zeros(out.shape, np.uint8)
cv2.fillPoly(m, [np.round(pts).astype(np.int32)], 1)
m = m.astype(bool)
if z["kind"] == "wall":
out[m & (out >= 4) & (out <= 9)] = 2 # a detection removed: it is ordinary painted wall
else:
out[m] = CORRECT_CLASS[z["kind"]]
return out
def objects(self, job_id):
"""Detected windows, doors, sockets and other protected objects as polygons (photo pixels), after the
operator's corrections, so the console can show them and let the operator say a detection is wrong."""
import cv2
from demo_plan import CLASS_IDS, REGION_STYLE, refine_classes
path = self.job_dir(job_id)
classes = np.asarray(Image.open(path / "classes.png"))
classes = refine_classes(self.corrected_classes(classes, self.get_zones(job_id)["zones"]))
names = {v: k for k, v in CLASS_IDS.items()}
out = []
for cid, name in sorted(names.items()):
mask = (classes == cid).astype(np.uint8)
contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
for c in sorted(contours, key=cv2.contourArea, reverse=True)[:60]:
area = float(cv2.contourArea(c))
if area < 30:
continue
poly = cv2.approxPolyDP(c, max(1.5, 0.008 * cv2.arcLength(c, True)), True).reshape(-1, 2)
if len(poly) < 3:
continue
out.append({"id": f"{name}-{len(out)}", "class": name, "label": REGION_STYLE[name][0], "area_px": area,
"polygon": [[int(x), int(y)] for x, y in poly]})
return {"objects": out, "width": int(classes.shape[1]), "height": int(classes.shape[0])}
def floor(self, job_id):
"""Floor mask + wall/floor line for a job, computed on first request (off by default in the console) and cached
until the operator's zones change. Uses the wall model's class map (with corrections) to keep the floor below
the wall and off windows / doors."""
path = self.job_dir(job_id)
zf = path / "zones.json"
stamp = zf.stat().st_mtime if zf.is_file() else 0
cache = path / "floor.json"
if cache.is_file() and (path / "layer_floor.png").is_file():
data = json.loads(cache.read_text())
if data.get("zones_stamp") == stamp:
return data
import floor_detect
rgb = np.asarray(Image.open(path / "photo.jpg").convert("RGB"))
classes = None
if (path / "classes.png").is_file():
classes = self.corrected_classes(np.asarray(Image.open(path / "classes.png")), self.get_zones(job_id)["zones"])
with self.lock:
started = time.time()
mask, info = floor_detect.detect_floor(rgb, classes)
info["seconds"] = round(time.time() - started, 1)
layer_png(mask, FLOOR_COLOUR, 110).save(path / "layer_floor.png")
edge_png(mask, FLOOR_COLOUR).save(path / "layer_floor_edge.png")
info["zones_stamp"] = stamp
info["width"], info["height"] = int(mask.shape[1]), int(mask.shape[0])
cache.write_text(json.dumps(info))
return info
# ---------------------------------------------------------------- drone digital twin (drone_twin.py)
TWIN_DEFAULTS = {"tip": "517", "pressure_psi": 2000, "standoff_mm": 300, "boom_m": 0.5, "masked_openings": True,
"home": None, "a_en": None, "b_en": None, "ground_alt_m": 0.0}
def _twin_setup(self, path, request=None):
f = path / "dronetwin_setup.json"
setup = {**self.TWIN_DEFAULTS, **(json.loads(f.read_text()) if f.is_file() else {})}
if request:
for k in self.TWIN_DEFAULTS:
if k in request:
setup[k] = request[k]
for k in ("home", "a_en", "b_en"):
v = setup[k]
if v is not None:
setup[k] = [float(x) for x in v][: 3 if k == "home" else 2]
if len(setup[k]) < (2 if k != "home" else 2):
raise ValueError(f"{k} needs numbers")
setup["masked_openings"] = bool(setup["masked_openings"])
f.write_text(json.dumps(setup, indent=1))
return setup
def _twin_spray(self, setup):
import application
tip = application.SprayTip.parse(str(setup["tip"]))
sp = application.spray_settings(tip.flow_lps(float(setup["pressure_psi"])), tip.fan_angle_deg, float(setup["standoff_mm"]), 38, 0.38)
return {**sp, "tip": str(setup["tip"]), "pressure_psi": float(setup["pressure_psi"])} # the audit needs both for the spray force
def _scale_mm_per_px(self, path):
saved = json.loads((path / "scale.json").read_text()) if (path / "scale.json").is_file() else {}
if saved.get("mm_per_px"):
return float(saved["mm_per_px"]), saved.get("source")
prediction = json.loads((path / "result.json").read_text())
mm, source = scale_from_prediction(prediction)
if not mm:
raise ValueError("this job has no scale yet: set one from a drawing, or plan it first")
return float(mm), source
def dronetwin(self, job_id, request=None, create=True):
"""The job's digital twin: built from the photo's class map (operator corrections applied) on first use,
saved in the job folder, and kept in step with later zone corrections. Returns (twin, setup, spray)."""
import drone_twin
path = self.job_dir(job_id)
setup = self._twin_setup(path, request)
spray = self._twin_spray(setup)
tf = path / "dronetwin.json"
classes = self.corrected_classes(np.asarray(Image.open(path / "classes.png")), self.get_zones(job_id)["zones"])
mm, source = self._scale_mm_per_px(path)
h, w = classes.shape
a_en = tuple(setup["a_en"]) if setup["a_en"] else (0.0, 0.0)
b_en = tuple(setup["b_en"]) if setup["b_en"] else (w * mm / 1000.0, 0.0)
if tf.is_file() and not (request and any(k in request for k in ("a_en", "b_en", "masked_openings"))):
twin = drone_twin.Twin.from_json(tf.read_text())
fac = twin.facades["wall"]
if fac.classes.shape == classes.shape and not np.array_equal(fac.classes, classes):
changed = fac.classes != classes
fac.classes = classes.copy()
twin._log("classes_updated", "wall", {"pixels_changed": int(changed.sum())})
fac.mm_per_px = mm
elif create:
twin = drone_twin.Twin(job_id)
twin.add_facade("wall", classes, mm, a_en, b_en, bool(setup["masked_openings"]))
else:
raise KeyError(job_id)
twin.scale_source = source
return twin, setup, spray
def save_dronetwin(self, job_id, twin):
path = self.job_dir(job_id)
(path / "dronetwin.json").write_text(twin.to_json())
from PIL import Image as _I
_I.fromarray(twin.heatmap("wall")).save(path / "dronetwin_heatmap.png")
(path / "dronetwin_geojson.json").write_text(json.dumps(twin.geojson()))
twin._loaded_revision = twin.revision # the files now match this revision
def dronetwin_status(self, job_id, request=None):
twin, setup, spray = self.dronetwin(job_id, request)
path = self.job_dir(job_id)
if getattr(twin, "_loaded_revision", None) != twin.revision or not (path / "dronetwin_heatmap.png").is_file():
self.save_dronetwin(job_id, twin) # write only when the twin changed (or has no files yet)
st = twin.status("wall", spray["fan_width_mm"])
fac = twin.facades["wall"]
return {"status": st, "setup": setup, "spray": {k: spray[k] for k in ("fan_width_mm", "pass_spacing_mm", "traverse_speed_mm_s")},
"facade": {"width_m": round(fac.classes.shape[1] * fac.mm_per_px / 1000.0, 2), "height_m": round(fac.classes.shape[0] * fac.mm_per_px / 1000.0, 2),
"mm_per_px": round(fac.mm_per_px, 3), "scale_source": getattr(twin, "scale_source", None)},
"calibration": twin.calibration, "events": twin.events[-12:], "sorties": sum(1 for e in twin.events if e["kind"] == "sortie"),
"history": twin.history("wall"), "observations": sum(1 for e in twin.events if e["kind"] == "observe"),
"spray_audit": twin.last_audit("wall"), "revision": twin.revision,
"width": int(fac.classes.shape[1]), "height": int(fac.classes.shape[0])}
def dronetwin_observe(self, job_id, data):
"""A re-scan photo of the wall: segment it, line it up with the job photo, fuse its paint read into the belief.
Pixels the new photo does not show, or does not call wall, are left unobserved (never assumed bare)."""
import drone_align
from PIL import ImageOps
twin, setup, spray = self.dronetwin(job_id)
try:
img = ImageOps.exif_transpose(Image.open(io.BytesIO(data))).convert("RGB")
except Exception as exc:
raise ValueError(f"not a readable photo ({exc})") from exc
if max(img.size) > MAX_SIDE:
img.thumbnail((MAX_SIDE, MAX_SIDE), Image.Resampling.LANCZOS)
new_rgb = np.asarray(img)
job_rgb = np.asarray(Image.open(self.job_dir(job_id) / "photo.jpg").convert("RGB"))
H, quality = drone_align.homography(job_rgb, new_rgb)
with self.lock:
_, masks = self.analyze_fn(new_rgb)
fac = twin.facades["wall"]
warped, covered = drone_align.warp_classes(np.asarray(masks["mask"], np.uint8), H, fac.classes.shape)
valid = covered & np.isin(warped, (1, 2, 3)) & fac.target
if valid.mean() < 0.02:
raise ValueError("the re-scan shows almost none of the wall to be painted: frame the same wall as the job photo")
twin.observe("wall", warped == 2, valid=valid)
self.save_dronetwin(job_id, twin)
return {"alignment": quality, "observed_share_of_target": round(float(valid.sum() / max(fac.target.sum(), 1)), 3),
**self.dronetwin_status(job_id)}
def dronetwin_next(self, job_id):
"""The next sortie: passes over what the belief says is still bare, as photo-pixel strokes (for the overlay) and as
a QGroundControl mission (needs the launch point). None when nothing worth a flight is left.
Fail-closed: when the plan's gate refuses the flight (wind, positioning, standoff, ceiling), the gate reasons are
returned and **no mission is emitted**, so the console has nothing to download. The passes are still returned so
the operator can see what the sortie would have done.
"""
import drone_plan
twin, setup, spray = self.dronetwin(job_id)
fac = twin.facades["wall"]
nxt = twin.next_sortie("wall", spray["fan_width_mm"])
if nxt is None:
return {"strokes": [], "mission": None, "note": "nothing left that a sortie would cover"}
strokes = nxt["plan"]["strokes"]
h = fac.classes.shape[0]
out = {"strokes": strokes, "passes": len(strokes), "mission": None}
if setup["home"]:
mm = fac.mm_per_px
strokes_mm = [[(x * mm, (h - y) * mm) for x, y in s] for s in strokes]
width_mm, height_mm = fac.classes.shape[1] * mm, h * mm
home = tuple(setup["home"]) + ((0.0,) if len(setup["home"]) == 2 else ())
plan = drone_plan.build(strokes_mm, width_mm, height_mm, fac.a_en, fac.b_en, home, spray,
standoff_m=float(setup["standoff_mm"]) / 1000.0, boom_m=float(setup["boom_m"]))
out["gate"] = plan["gate"]
out["sorties"] = plan["sorties"]
out["reaction"] = plan.get("reaction")
if plan["gate"]["go"]:
along, outward = plan["_frame"]
out["mission"] = drone_plan.qgc_mission(plan["_sorties"], home, fac.a_en, along, outward,
plan["standoff_setpoint_mm"] / 1000.0 + float(setup["boom_m"]), plan["speed_ms"])
else:
out["refused"] = True
out["note"] = "flight plan withheld: " + "; ".join(plan["gate"]["reasons"])
return out
def dronetwin_whatif(self, job_id, positioning_m=0.02, wind_ms=2.0, trials=3):
"""What this facade would be predicted to get under a different position error and wind.
The twin's own answer from its plan and deposit model - a prediction, not a measurement - so the operator can see
what a windier day or a sloppier hold costs before flying it.
"""
twin, setup, spray = self.dronetwin(job_id)
positioning_m = min(max(float(positioning_m), 0.002), 0.30)
wind_ms = min(max(float(wind_ms), 0.0), 12.0)
trials = max(1, min(int(trials), 6))
out = twin.what_if("wall", spray, positioning_m, wind_ms, trials=trials)
out.update({"positioning_m": positioning_m, "wind_ms": wind_ms, "trials": trials,
"masked_openings": bool(twin.facades["wall"].masked_openings)})
return out
def dronetwin_log(self, job_id, text, sortie=None):
"""Ingest a flight log (CSV) into the job's twin; returns what the flight did and where the facade stands."""
twin, setup, spray = self.dronetwin(job_id)
if not setup["home"]:
raise ValueError("set the home position first (lat, lon): the facade's east/north coordinates are measured from it")
result = twin.ingest_log("wall", text, spray["fan_width_mm"], spray["pass_spacing_mm"], target_standoff_m=float(setup["standoff_mm"]) / 1000.0,
boom_m=float(setup["boom_m"]), home=tuple(setup["home"]), ground_alt_m=float(setup["ground_alt_m"]), sortie=sortie,
spray=spray)
self.save_dronetwin(job_id, twin)
return {"ingest": result, **self.dronetwin_status(job_id)}
# ---------------------------------------------------------------- job drawings
def _drawings(self, job_id):
path = self.job_dir(job_id) / "drawings"
index = path / "index.json"
return path, (json.loads(index.read_text()) if index.is_file() else [])
def add_drawing(self, job_id, data, filename=""):
"""Attach a plan / elevation to a job: an image, or the first page of a PDF, kept as a PNG to click on."""
if len(data) > MAX_DRAWING:
raise ValueError(f"drawing larger than {MAX_DRAWING // (1024 * 1024)} MB")
path, items = self._drawings(job_id)
ext = Path(filename).suffix.lower()
pages = 1
if data[:5] == b"%PDF-" or ext == ".pdf":
try:
import pypdfium2 as pdfium
pdf = pdfium.PdfDocument(data)
pages = len(pdf)
page = pdf[0]
w, h = page.get_size()
image = page.render(scale=min(4.0, DRAWING_SIDE / max(w, h))).to_pil().convert("RGB")
except ImportError as exc:
raise ValueError("PDF drawings need pypdfium2 installed on the server; upload a PNG or JPG instead") from exc
except Exception as exc:
raise ValueError(f"could not read that PDF: {exc}") from exc
else:
try:
image = Image.open(io.BytesIO(data))
image.load()
except Exception as exc:
raise ValueError(f"not a readable drawing (use PNG, JPG or PDF): {exc}") from exc
from PIL import ImageOps
image = ImageOps.exif_transpose(image).convert("RGB")
if max(image.size) > DRAWING_SIDE:
image.thumbnail((DRAWING_SIDE, DRAWING_SIDE), Image.Resampling.LANCZOS)
path.mkdir(exist_ok=True)
did = "d" + secrets.token_hex(4)
image.save(path / f"{did}.png")
item = {"id": did, "filename": (filename or "drawing")[:120], "width": image.width, "height": image.height,
"pages": pages, "created": now(), "calibration": None, "measurements": []}
items.append(item)
(path / "index.json").write_text(json.dumps(items, indent=1))
return item
def list_drawings(self, job_id):
return self._drawings(job_id)[1]
def update_drawing(self, job_id, drawing_id, request):
"""Store the operator's calibration (two clicked points + the real length) and measured lines."""
path, items = self._drawings(job_id)
item = next((d for d in items if d["id"] == drawing_id), None)
if item is None:
raise KeyError(drawing_id)
if "calibration" in request:
c = request["calibration"]
if c is not None:
mm = float(c["length_mm"])
px = float(c["length_px"])
if not (mm > 0 and px > 0):
raise ValueError("a calibration needs a positive length and two different points")
c = {"length_mm": mm, "length_px": px, "mm_per_px": mm / px, "label": str(c.get("label", ""))[:40],
"pts": _points(c.get("pts"))}
item["calibration"] = c
if "measurements" in request:
item["measurements"] = [{"id": str(m.get("id", ""))[:12], "length_px": float(m["length_px"]),
"label": str(m.get("label", ""))[:60], "pts": _points(m.get("pts"))}
for m in request["measurements"]][:60]
(path / "index.json").write_text(json.dumps(items, indent=1))
return item
def drawing_file(self, job_id, drawing_id):
if not DRAWING_ID.match(drawing_id):
raise KeyError(drawing_id)
f = self.job_dir(job_id) / "drawings" / f"{drawing_id}.png"
if not f.is_file():
raise KeyError(drawing_id)
return f
def set_scale_from_drawing(self, job_id, request):
"""A length measured on a drawing (the wall's width or height) fixes the photo's scale: the wall's extent in
the photo's pixels is known from its mask, so mm per pixel follows. Saved for every later plan of this job."""
path = self.job_dir(job_id)
if json.loads((path / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to change the scale")
length_mm = float(request["length_mm"])
axis = request.get("axis")
if axis not in ("width", "height") or not 100 <= length_mm <= 60000:
raise ValueError("give the wall's width or height, between 0.1 m and 60 m")
wall = load_mask(path / "wall.png")
ys, xs = np.nonzero(wall)
if not len(xs):
raise ValueError("no wall was found in the photo, so there is nothing to scale")
extent = (xs.max() - xs.min() + 1) if axis == "width" else (ys.max() - ys.min() + 1)
operator = request.get("source") == "operator" # a length the operator measured on the real wall, not read off a drawing
saved = {"mm_per_px": length_mm / float(extent),
"source": (f"operator measured: wall {axis} = {length_mm / 1000:.2f} m" if operator else
f"drawing: {str(request.get('label') or axis)[:60]} = {length_mm / 1000:.2f} m ({axis})"),
"axis": axis, "length_mm": length_mm, "at": now()}
(path / "scale.json").write_text(json.dumps(saved, indent=1))
self._record_anchor(path, ("measured_" + axis) if operator else "drawing", saved["mm_per_px"], saved["source"])
saved["evidence"] = self.evidence(job_id)
return saved
def generate_drawing(self, job_id, request):
"""Draw this job's wall to scale from its photo (drawing_make.py) and file it with the job's drawings, ready to calibrate-check,
match, measure and download as SVG, DXF and JSON. The scale comes from, in order: a wall height or width the operator gives, the
job's saved scale, or an assumed 8 ft ceiling, and the sheet says which in its title block. It is a DRAFT."""
import cv2
import drawing_make as dmk
from wall_measure import measure_wall
path, items = self._drawings(job_id)
job = self.job_dir(job_id)
classes = job / "classes.png"
if not classes.is_file():
raise ValueError("this job has no saved class mask; re-run the photo through the current model")
labels = np.asarray(Image.open(classes))
kw = {}
height_ft, width_ft = request.get("wall_height_ft"), request.get("wall_width_ft")
if height_ft not in (None, ""):
height_ft = float(height_ft)
if not 4 <= height_ft <= 40:
raise ValueError("wall height must be between 4 and 40 ft")
kw["wall_height_ft"] = height_ft
if width_ft not in (None, ""):
width_ft = float(width_ft)
if not 1 <= width_ft <= 200:
raise ValueError("wall width must be between 1 and 200 ft")
kw["wall_width_ft"] = width_ft
explicit = bool(kw) # an height or width typed for THIS drawing means "draw from the photo with that size"
scale = job / "scale.json"
if not kw and scale.is_file():
kw["mm_per_px"] = float(json.loads(scale.read_text())["mm_per_px"])
info = json.loads((job / "job.json").read_text())
spec = None
if request.get("from_twin", True) and not explicit and (job / "twin.npz").is_file(): # a twin's whole-wall size and fused openings beat one photo
_, twin = self._twin_load(job_id)
spec = dmk.make_elevation_from_twin(twin, info.get("wall_id") or info.get("filename") or "Wall")
if spec is None:
measured = measure_wall(labels, find_corners=False, **kw)
if not measured["faces"]:
raise ValueError("no measurable wall in this photo, so there is nothing to draw")
face = max(measured["faces"], key=lambda f: f["width_ft"] * f["height_ft"])
spec = dmk.make_elevation(labels, face, info.get("wall_id") or info.get("filename") or "Wall",
source_note="drawn from one photo" + (f" ({len(measured['faces'])} wall faces seen; the largest is drawn)" if len(measured["faces"]) > 1 else ""))
img, mm_per_px = dmk.to_png(spec)
did = "d" + secrets.token_hex(4)
path.mkdir(exist_ok=True)
cv2.imwrite(str(path / f"{did}.png"), img)
(path / f"{did}.json").write_text(dmk.dumps(spec))
(path / f"{did}.svg").write_text(dmk.to_svg(spec))
(path / f"{did}.dxf").write_text(dmk.to_dxf(spec))
item = {"id": did, "filename": f"{spec['name']} - generated elevation (draft).png", "width": int(img.shape[1]), "height": int(img.shape[0]),
"pages": 1, "created": now(), "generated": True,
"calibration": {"length_mm": 100 * mm_per_px, "length_px": 100.0, "mm_per_px": mm_per_px, "label": "generated sheet",
"pts": None}, # no clicked points: nothing to draw over the sheet
"measurements": [], "provenance": spec["provenance"], "openings": len(spec["openings"]), "schedule": dmk.schedule(spec),
"wall_mm": [spec["wall"]["width_mm"], spec["wall"]["height_mm"]]}
items.append(item)
(path / "index.json").write_text(json.dumps(items, indent=1))
return item
def drawing_export(self, job_id, drawing_id, ext):
if not DRAWING_ID.match(drawing_id) or ext not in ("svg", "dxf", "json"):
raise KeyError(drawing_id)
f = self.job_dir(job_id) / "drawings" / f"{drawing_id}.{ext}"
if not f.is_file():
raise KeyError(drawing_id)
return f
# ------------------------------------------------------------------ evidence: what supports this job's measurement (evidence.py)
def _record_anchor(self, path, kind, mm_per_px, label):
"""Remember every independent way the scale was fixed (so agreement or conflict between them can be shown), latest per kind."""
f = path / "anchors.json"
anchors = json.loads(f.read_text()) if f.is_file() else {}
anchors[kind] = {"mm_per_px": float(mm_per_px), "label": str(label)[:120], "at": now()}
f.write_text(json.dumps(anchors, indent=1))
def _views(self, path):
f = path / "views.json"
return json.loads(f.read_text()) if f.is_file() else []
def _main_arrays(self, path):
import cv2
gray = cv2.imread(str(path / "photo.jpg"), cv2.IMREAD_GRAYSCALE)
return gray, np.asarray(Image.open(path / "classes.png"))
def _refresh_coverage(self, path):
import evidence_inputs as ei
gray, labels = self._main_arrays(path)
regs = []
for v in self._views(path):
if v.get("registered") and (path / "views" / f"{v['id']}.jpg").is_file():
import cv2
regs.append((cv2.imread(str(path / "views" / f"{v['id']}.jpg"), cv2.IMREAD_GRAYSCALE), np.asarray(Image.open(path / "views" / f"{v['id']}_classes.png"))))
cov = ei.coverage(gray, labels, regs)
(path / "coverage.json").write_text(json.dumps(cov, indent=1))
return cov
def _add_one_view(self, path, rgb, kind, name):
"""Segment one extra image, try to register it to the job's own photo, record the result. Never raises for a view that fails to register."""
import cv2
import evidence_inputs as ei
views = self._views(path)
if len(views) >= MAX_VIEWS:
raise ValueError(f"at most {MAX_VIEWS} extra views per wall")
gray_main, labels_main = self._main_arrays(path)
h, w = gray_main.shape
rgb = cv2.resize(np.asarray(rgb), (w, h), interpolation=cv2.INTER_AREA)
with self.lock:
summary, masks = self.analyze_fn(rgb)
labels = np.asarray(masks["mask"], np.uint8) if masks.get("mask") is not None else None
if labels is None:
raise ValueError("the model returned no class mask for this view")
gray = cv2.cvtColor(rgb, cv2.COLOR_RGB2GRAY)
counted = [cv2.imread(str(path / "views" / f"{v['id']}.jpg"), cv2.IMREAD_GRAYSCALE) for v in views
if v.get("registered") and (path / "views" / f"{v['id']}.jpg").is_file()]
ok, inliers, why = ei.register_view(gray_main, labels_main, gray, labels, existing=counted)
vid = "v" + secrets.token_hex(4)
rec = {"id": vid, "kind": kind, "name": str(name)[:100], "registered": bool(ok), "inliers": int(inliers), "reason": why,
"wall_pixels": int(summary.get("wall_pixels", 0)), "at": now()}
if ok:
(path / "views").mkdir(exist_ok=True)
cv2.imwrite(str(path / "views" / f"{vid}.jpg"), cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR), [cv2.IMWRITE_JPEG_QUALITY, 90])
Image.fromarray(labels).save(path / "views" / f"{vid}_classes.png")
views.append(rec)
(path / "views.json").write_text(json.dumps(views, indent=1))
return rec
def add_view(self, job_id, data, filename=""):
"""Another photo or a video of the same wall. Each image must register to the wall's own photo to count; a video contributes up to six
sharp, distinct keyframes (each counts half, they are correlated). The result lists what registered and what did not, and why."""
import cv2
from multiview import video_keyframes
path = self.job_dir(job_id)
if json.loads((path / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to add evidence")
ext = Path(filename).suffix.lower().lstrip(".")
added = []
if ext in VIDEO_EXT:
if len(data) > MAX_VIDEO:
raise ValueError(f"video larger than {MAX_VIDEO // (1024 * 1024)} MB")
tmp = path / f"_upload.{ext}"
tmp.write_bytes(data)
try:
frames = video_keyframes(tmp, every_sec=1.0, max_frames=6)
finally:
tmp.unlink(missing_ok=True)
if len(frames) < 1:
raise ValueError("no sharp frames could be read from that video")
for i, bgr in enumerate(frames):
added.append(self._add_one_view(path, cv2.cvtColor(bgr, cv2.COLOR_BGR2RGB), "keyframe", f"{filename or 'video'}#{i}"))
else:
if len(data) > MAX_UPLOAD:
raise ValueError(f"image larger than {MAX_UPLOAD // (1024 * 1024)} MB")
try:
image = Image.open(io.BytesIO(data))
image.load()
except Exception as exc:
raise ValueError(f"not a readable photo or video: {exc}") from exc
from PIL import ImageOps
image = ImageOps.exif_transpose(image).convert("RGB")
if max(image.size) > MAX_SIDE:
image.thumbnail((MAX_SIDE, MAX_SIDE), Image.Resampling.LANCZOS)
added.append(self._add_one_view(path, np.asarray(image), "photo", filename or "photo"))
cov = self._refresh_coverage(path)
return {"added": added, "registered": sum(1 for a in added if a["registered"]), "tried": len(added), "coverage": cov,
"evidence": self.evidence(job_id)}
def add_depth(self, job_id, data, filename="", fx=None, hfov_deg=None):
"""A depth / LiDAR map aligned with the job's photo: checked against the wall mask, and its metric scale is recorded as a scale anchor.
The focal length (fx in depth-image pixels, or the horizontal field of view) decides whether that scale is trusted."""
import evidence_inputs as ei
path = self.job_dir(job_id)
if json.loads((path / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to add evidence")
if len(data) > MAX_DEPTH:
raise ValueError(f"depth map larger than {MAX_DEPTH // (1024 * 1024)} MB")
depth = ei.load_depth_bytes(data, filename)
wall, keep = load_mask(path / "wall.png"), load_mask(path / "keepout.png")
job = json.loads((path / "job.json").read_text())
rep = ei.depth_check(depth, wall, keep, (job["width"], job["height"]), fx=fx, hfov_deg=hfov_deg)
rep["filename"], rep["at"] = Path(filename).name[:120], now()
(path / "depth_report.json").write_text(json.dumps(rep, indent=1))
if rep["plane_ok"]: # the relief (mm from the wall plane) per photo pixel feeds the twin's relief layer
import twin_sources as ts
resid, valid, rinfo = ts.relief_from_depth(depth, (job["width"], job["height"]), fx_px=fx, hfov_deg=hfov_deg)
np.savez_compressed(path / "depth_relief.npz", resid=resid.astype(np.float16), valid=valid)
if rep.get("mm_per_px") and rep["plane_ok"]:
kind = "depth" if rep["intrinsics_source"] == "known" else "assumed_focal"
self._record_anchor(path, kind, rep["mm_per_px"], f"depth scan ({rep['intrinsics_source']} focal length)")
(path / "scale.json").write_text(json.dumps({"mm_per_px": rep["mm_per_px"], "source": f"depth {rep['intrinsics_source']} focal", "axis": "depth",
"length_mm": None, "at": now()}, indent=1))
return {"report": {k: rep.get(k) for k in ("plane_ok", "wall_fraction_on_plane", "mm_per_px", "intrinsics_source", "fx_px", "warnings", "points")},
"evidence": self.evidence(job_id)}
def evidence(self, job_id):
"""The evidence score for this job and what would raise it, from what is recorded on disk (evidence.py explains the rules)."""
import evidence
path = self.job_dir(job_id)
views = self._views(path)
photos = [v for v in views if v["kind"] == "photo" and v["registered"]]
keyframes = [v for v in views if v["kind"] == "keyframe" and v["registered"]]
cov = json.loads((path / "coverage.json").read_text()) if (path / "coverage.json").is_file() else self._refresh_coverage(path)
anchors = json.loads((path / "anchors.json").read_text()) if (path / "anchors.json").is_file() else {}
facts = {"views": 1 + len(photos), "video_keyframes": len(keyframes), "wall_complete": bool(cov.get("complete")), "scale": "assumed"}
if anchors:
facts["anchors"] = [{"kind": k, "mm_per_px": a["mm_per_px"]} for k, a in anchors.items()]
drawings = [d for d in self._drawings(job_id)[1] if not d.get("generated")] # a drawing made FROM the photo is not independent evidence
if any((d.get("auto_match") or {}).get("usable_for_scale") for d in drawings):
facts["drawing"] = "matched"
elif any(d.get("calibration") for d in drawings):
facts["drawing"] = "calibrated"
dr = json.loads((path / "depth_report.json").read_text()) if (path / "depth_report.json").is_file() else None
facts["depth_plane_ok"] = bool(dr and dr.get("plane_ok"))
out = evidence.assess(facts)
out["facts"] = facts
out["views"] = [{k: v.get(k) for k in ("id", "kind", "name", "registered", "inliers", "reason")} for v in views]
out["depth"] = None if not dr else {k: dr.get(k) for k in ("plane_ok", "wall_fraction_on_plane", "mm_per_px", "intrinsics_source", "warnings", "filename")}
out["coverage"] = cov
return out
# ------------------------------------------------------------------ live drone feed (live_feed.py): advisory only, one session at a time
def list_videos(self):
"""Uploaded videos (newest first): what the live panel can replay as a feed."""
out = []
for d in sorted(self.jobs.iterdir(), key=lambda p: p.stat().st_mtime, reverse=True):
if VIDEO_ID.match(d.name) and any(d.glob("video.*")):
out.append({"id": d.name, "modified": time.strftime("%Y-%m-%d %H:%M", time.localtime(d.stat().st_mtime))})
return out[:30]
def _person_detector(self):
"""The person detector this machine can run, built once and reused. None means not configured - never 'clear'."""
if not hasattr(self, "_people_det"):
import people
try:
self._people_det = people.make_detector()
except Exception:
self._people_det = None
return self._people_det
def live_start(self, request):
"""Open a camera URL (rtsp/http), or play one of this console's own uploaded videos as if it were live. Replaces any running session."""
import live_feed as lf
self.live_stop()
if request.get("video_id"):
vid = str(request["video_id"])
if not VIDEO_ID.match(vid):
raise ValueError("not a video id")
clip = next((p for p in sorted((self.jobs / vid).glob("video.*")) if p.suffix.lstrip(".") in VIDEO_EXT), None) if (self.jobs / vid).is_dir() else None
if clip is None:
raise KeyError(vid)
grabber, label = lf.CaptureGrabber(str(clip), realtime=True, loop=True), f"recorded video {vid} (played as live)"
else:
url = lf.parse_source(request.get("source"))
grabber, label = lf.CaptureGrabber(url), "camera stream"
def analyze(rgb):
with self.lock:
return self.analyze_fn(rgb)
twin_job = request.get("twin_job")
sink = None
if twin_job:
if not JOB_ID.match(str(twin_job)):
raise ValueError("not a job id")
self._twin_load(twin_job) # raises KeyError (404) when that job has no twin yet
sink = lambda rgb, masks, summary: self.twin_observe_live(twin_job, rgb, masks, summary) # noqa: E731
session = lf.LiveSession(analyze, grabber, interval_s=1.0, on_frame=sink,
people_fn=self._person_detector() if request.get("people", True) else None)
session.start()
self.live = {"session": session, "label": label, "started": now(), "twin_job": twin_job}
return self.live_state()
def live_state(self):
if not getattr(self, "live", None):
return {"running": False, "disclaimer": None, "people_detector": getattr(self._person_detector(), "name", "none")}
out = self.live["session"].snapshot()
out["source"], out["started"] = self.live["label"], self.live["started"]
out["people_detector"] = getattr(getattr(self, "live", {}).get("session").people_fn, "name", "none") or "none"
return out
def live_frame(self):
live = getattr(self, "live", None)
jpg = live["session"].annotated if live else None
if not jpg:
raise KeyError("no live frame yet")
return jpg
def live_arm(self, request):
live = getattr(self, "live", None)
if not live:
raise ValueError("no live session is running")
if request.get("armed"):
live["session"].arm(request.get("operator", ""))
else:
live["session"].disarm()
return self.live_state()
def live_stop(self):
live = getattr(self, "live", None)
if live:
live["session"].stop()
self.live = None
return {"running": False}
# ------------------------------------------------------------------ digital twin (twin.py): a per-job wall model fed by every observation
TWIN_LAYERS = ("painted", "confidence", "plan", "surface", "relief")
def _twin_clock(self, path):
"""Observation time in seconds since this twin was built: a monotonic clock that survives restarts (stored as the build time)."""
m = json.loads((path / "twin_meta.json").read_text())
return max(0.0, time.time() - float(m["epoch"]))
def _anchor_from_dims(self, path, kind, dims_mm, label):
"""Record a laser-derived size as a scale anchor, in the SAME definition as the console's other anchors (mm per pixel = a real length over the
wall mask's extent in the photo), and make it the job's scale. Returns mm per pixel."""
from wall_measure import measure_wall
wall = load_mask(path / "wall.png")
ys, xs = np.nonzero(wall)
if not len(xs):
raise ValueError("no wall was found in the photo, so a size cannot be tied to it")
faces = measure_wall(np.asarray(Image.open(path / "classes.png")), find_corners=False)["faces"] if (path / "classes.png").is_file() else []
if faces and {"top", "bottom"} & set(max(faces, key=lambda f: f["width_ft"] * f["height_ft"])["clipped_sides"]):
return None # the photo cuts the wall off vertically: its pixel height is not the wall's height, so it cannot anchor a scale
extent_h = float(ys.max() - ys.min() + 1)
mm_per_px = float(dims_mm[1]) / extent_h
self._record_anchor(path, kind, mm_per_px, label)
(path / "scale.json").write_text(json.dumps({"mm_per_px": mm_per_px, "source": label, "axis": "height", "length_mm": float(dims_mm[1]), "at": now()}, indent=1))
return mm_per_px
def add_laser(self, job_id, request):
"""Laser distance meter readings of the wall (width, height, optionally the diagonal): validated, stored, and used as the twin's size on the next build."""
import twin_sources as ts
path = self.job_dir(job_id)
if json.loads((path / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to add evidence")
got = ts.laser_dimensions(request.get("width_mm"), request.get("height_mm"), request.get("diagonal_mm"), request.get("accuracy_mm"))
rec = {"width_mm": got["dims_mm"][0], "height_mm": got["dims_mm"][1], "accuracy_mm": got["accuracy_mm"], "diagonal_check": got["diagonal_check"], "at": now()}
(path / "laser.json").write_text(json.dumps(rec, indent=1))
anchored = self._anchor_from_dims(path, "laser", got["dims_mm"], f"laser distance meter: {got['dims_mm'][0] / 1000:.3f} x {got['dims_mm'][1] / 1000:.3f} m")
return {"laser": rec, "scale_anchored": anchored is not None, "evidence": self.evidence(job_id),
"note": "used as the twin's size on the axes the photos show end to end, the next time it is built" + ("" if anchored else "; the photo cuts the wall off vertically, so it cannot set the photo's scale yet")}
def add_scan(self, job_id, data, filename="", up_axis="z", flip_x=False):
"""A laser-scan point cloud (PLY / XYZ / CSV, metres) of the wall: plane-fitted for the wall's size and its relief. The scan's vertical axis and
left-right order are the operator's statement (the cloud's frame is arbitrary) and the result repeats that assumption."""
import twin_sources as ts
path = self.job_dir(job_id)
if json.loads((path / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to add evidence")
if not data or len(data) > MAX_SCAN:
raise ValueError(f"a scan must be between 1 byte and {MAX_SCAN // (1024 * 1024)} MB")
pts = ts.parse_point_cloud(data, filename)
fit = ts.fit_wall_from_cloud(pts, up_axis=up_axis, flip_x=flip_x)
meta = {k: fit[k] for k in ("dims_mm", "plane_rms_mm", "inlier_fraction", "points", "uncertainty_mm", "warnings", "alignment", "normal")}
meta.update(filename=Path(filename).name[:120], at=now())
np.savez_compressed(path / "scan.npz", relief=np.nan_to_num(fit["relief_mm"]).astype(np.float16), weight=fit["relief_w"].astype(np.float16),
meta=np.frombuffer(json.dumps(meta).encode(), np.uint8))
anchored = self._anchor_from_dims(path, "laser_scan", fit["dims_mm"], f"laser scan {Path(filename).name[:40]}: {fit['dims_mm'][0] / 1000:.3f} x {fit['dims_mm'][1] / 1000:.3f} m")
return {"scan": meta, "relief_cells": int((fit["relief_w"] > 0).sum()), "scale_anchored": anchored is not None, "evidence": self.evidence(job_id),
"note": "used as the twin's size (on axes the photos show end to end) and relief the next time it is built"}
def _scan_load(self, path):
f = path / "scan.npz"
if not f.is_file():
return None
z = np.load(f, allow_pickle=False)
return {"relief": z["relief"].astype(np.float32), "weight": z["weight"].astype(np.float32), "meta": json.loads(bytes(z["meta"]).decode())}
def twin_build(self, job_id, request):
"""Build (or rebuild) this job's twin from ALL the evidence on file, deterministically: the main photo, every registered extra view and video
keyframe (fused, so the wall is measured whole), laser distance readings, a laser scan, and a depth map's relief. The wall's SIZE comes from the most
accurate source present (scan, then laser meter, then photo geometry) and the others are reported against it, never averaged into it."""
import cv2
import evidence_inputs as ei
import twin as tn
import twin_sources as ts
from wall_measure import measure_wall
path = self.job_dir(job_id)
if not (path / "classes.png").is_file():
raise ValueError("this job has no saved class mask; re-run the photo through the current model")
gray_main, labels = self._main_arrays(path)
kw = {}
for key in ("wall_height_ft", "wall_width_ft"):
v = request.get(key)
if v not in (None, ""):
kw[key] = float(v)
laser = json.loads((path / "laser.json").read_text()) if (path / "laser.json").is_file() else None
scan = self._scan_load(path)
if not kw and (path / "scale.json").is_file() and not (laser or scan):
kw["mm_per_px"] = float(json.loads((path / "scale.json").read_text())["mm_per_px"])
# extra views that register to the main photo, with their homographies
regs = []
if request.get("fuse_views", True):
for v in self._views(path):
f = path / "views" / f"{v['id']}.jpg"
if v.get("registered") and f.is_file():
g = cv2.imread(str(f), cv2.IMREAD_GRAYSCALE)
lab = np.asarray(Image.open(path / "views" / f"{v['id']}_classes.png"))
H, inl, why = ei.register_homography(gray_main, labels, g, lab)
if H is not None:
regs.append((v, g, lab, H))
if regs:
face, finfo = ts.fuse_views_face(gray_main, labels, [(g, lab) for _, g, lab, _ in regs], kw)
else:
faces = measure_wall(labels, find_corners=False, **kw)["faces"]
if not faces:
raise ValueError("no measurable wall in this photo, so there is nothing to build a twin of")
face = max(faces, key=lambda f: f["width_ft"] * f["height_ft"])
finfo = {"views_used": 1, "complete": not face["clipped_sides"], "clipped_sides": list(face["clipped_sides"])}
prov = {"job": job_id, "built": now(), "views_fused": finfo["views_used"], "wall_complete": finfo["complete"], "clipped_sides": finfo["clipped_sides"]}
dims = None
if laser:
prov["laser"] = {"accuracy_mm": laser["accuracy_mm"], "at": laser["at"]}
if laser.get("diagonal_check"):
prov["diagonal_check"] = laser["diagonal_check"]
dims = (laser["width_mm"], laser["height_mm"])
if scan:
sm = scan["meta"]
prov["laser_scan"] = {"uncertainty_mm": sm["uncertainty_mm"], "plane_rms_mm": sm["plane_rms_mm"], "inlier_fraction": sm["inlier_fraction"],
"file": sm.get("filename"), "alignment": sm["alignment"], "warnings": sm["warnings"], "note": sm["alignment"]["assumption"]}
if laser:
prov["scan_vs_laser_pct"] = {"width": round(100 * (sm["dims_mm"][0] - laser["width_mm"]) / laser["width_mm"], 2),
"height": round(100 * (sm["dims_mm"][1] - laser["height_mm"]) / laser["height_mm"], 2)}
dims = tuple(sm["dims_mm"]) # the scan is the most accurate source present
info = json.loads((path / "job.json").read_text())
twin = tn.WallTwin.from_face(labels, face, info.get("wall_id") or info.get("filename") or "Wall", cell_mm=float(request.get("cell_mm") or 50.0), t=0.0,
provenance=prov, dims_mm=dims)
if (path / "touchup.png").is_file(): # the photo's damage / touch-up regions, placed on the wall
tm = np.asarray(Image.open(path / "touchup.png").convert("L")) > 0
if tm.shape == labels.shape:
twin.damage = np.where(twin._to_grid(tm.astype(np.uint8), np.eye(3)) == 1, 1, twin.damage).astype(np.uint8)
observed = []
for k, (v, g, lab, H) in enumerate(regs, start=1):
twin.observe_view(lab, H, float(k), v.get("name") or v["id"])
observed.append(v["id"])
if scan:
twin.observe_relief_grid(scan["relief"], scan["weight"], 0.0, "laser scan", weight=3.0)
if (path / "depth_relief.npz").is_file():
z = np.load(path / "depth_relief.npz", allow_pickle=False)
twin.observe_relief_photo(z["resid"].astype(np.float32), z["valid"], np.eye(3), 0.0, "depth map", weight=1.0)
self._twin_set_plan(path, twin, labels.shape)
(path / "twin.npz").write_bytes(twin.to_bytes())
(path / "twin_meta.json").write_text(json.dumps({"epoch": time.time() - float(len(regs) + 1), "observed_views": observed, "built": now(),
"sources": {"views": len(regs), "laser": bool(laser), "scan": bool(scan), "depth_relief": (path / "depth_relief.npz").is_file()}}, indent=1))
return self.twin_state(job_id)
def twin_build_v1(self, job_id, request):
"""Build (or rebuild) this job's twin from its photo: the wall's size comes from the same measurement a generated drawing uses, so a measured
height gives true millimetres and an assumed one is flagged on the twin. The photo's planned strokes become the twin's plan."""
import cv2
import twin as tn
from wall_measure import measure_wall
path = self.job_dir(job_id)
if not (path / "classes.png").is_file():
raise ValueError("this job has no saved class mask; re-run the photo through the current model")
labels = np.asarray(Image.open(path / "classes.png"))
kw = {}
for key in ("wall_height_ft", "wall_width_ft"):
v = request.get(key)
if v not in (None, ""):
kw[key] = float(v)
if not kw and (path / "scale.json").is_file():
kw["mm_per_px"] = float(json.loads((path / "scale.json").read_text())["mm_per_px"])
faces = measure_wall(labels, find_corners=False, **kw)["faces"]
if not faces:
raise ValueError("no measurable wall in this photo, so there is nothing to build a twin of")
face = max(faces, key=lambda f: f["width_ft"] * f["height_ft"])
info = json.loads((path / "job.json").read_text())
twin = tn.WallTwin.from_face(labels, face, info.get("wall_id") or info.get("filename") or "Wall", cell_mm=float(request.get("cell_mm") or 50.0), t=0.0,
provenance={"job": job_id, "built": now()})
self._twin_set_plan(path, twin, labels.shape)
(path / "twin.npz").write_bytes(twin.to_bytes())
(path / "twin_meta.json").write_text(json.dumps({"epoch": time.time(), "observed_views": [], "built": now()}, indent=1))
return self.twin_state(job_id)
def _twin_set_plan(self, path, twin, shape):
import twin as tn
f = path / "work_order.json"
if not f.is_file():
return
steps = (json.loads(f.read_text()).get("steps") or [])
for st in steps:
if st.get("step") == "paint" and st.get("plan") and st["plan"].get("plan"):
plan = st["plan"]["plan"]
twin.set_plan(tn.rasterise_plan(plan.get("strokes"), float(plan.get("stroke_width_units") or 8), shape))
return
def _twin_load(self, job_id):
import twin as tn
path = self.job_dir(job_id)
if not (path / "twin.npz").is_file():
raise KeyError("no twin built for this job yet")
return path, tn.WallTwin.from_bytes((path / "twin.npz").read_bytes())
def twin_update(self, job_id):
"""Fold every registered extra view of this job into the twin (each once), and refresh the plan from the current work order."""
import cv2
import evidence_inputs as ei
path, twin = self._twin_load(job_id)
meta = json.loads((path / "twin_meta.json").read_text())
done = set(meta.get("observed_views", []))
gray_main, labels_main = self._main_arrays(path)
added = []
for v in self._views(path):
if not v.get("registered") or v["id"] in done:
continue
f = path / "views" / f"{v['id']}.jpg"
if not f.is_file():
continue
g = cv2.imread(str(f), cv2.IMREAD_GRAYSCALE)
lab = np.asarray(Image.open(path / "views" / f"{v['id']}_classes.png"))
H, inl, why = ei.register_homography(gray_main, labels_main, g, lab)
if H is None:
continue
twin.observe_view(lab, H, self._twin_clock(path), v.get("name") or v["id"])
done.add(v["id"]); added.append(v["id"])
self._twin_set_plan(path, twin, labels_main.shape)
(path / "twin.npz").write_bytes(twin.to_bytes())
meta["observed_views"] = sorted(done)
(path / "twin_meta.json").write_text(json.dumps(meta, indent=1))
out = self.twin_state(job_id)
out["added_views"] = added
return out
# ------------------------------------------------------------------ typed instructions (prompt_cmd.py): proposals first, applied on request
def prompt_context(self, job_id):
path = self.job_dir(job_id)
job = json.loads((path / "job.json").read_text())
wall = load_mask(path / "wall.png")
ys, xs = np.nonzero(wall)
box = [int(xs.min()), int(ys.min()), int(xs.max()), int(ys.max())] if len(xs) else None
return {"width": job["width"], "height": job["height"], "wall_box": box, "objects": self.objects(job_id)["objects"], "has_twin": (path / "twin.npz").is_file()}
def prompt(self, job_id, request, env=None):
"""Parse a typed instruction. Returns the proposed actions and the answers to any questions; applies NOTHING."""
import prompt_cmd as pc
ctx = self.prompt_context(job_id)
text = str(request.get("text") or "")
plan = pc.parse(text, ctx)
if not (plan["actions"] or plan["answers"] or plan["questions"] or plan["refused"]) and text.strip():
ai = pc.ai_parse(text, ctx, env)
if ai is not None and (ai["actions"] or ai["questions"]):
plan = ai
plan["actions"], dropped = pc.validate(plan["actions"], ctx)
plan["answered"] = [{"topic": a["topic"], "text": self.prompt_answer(job_id, a["topic"], ctx)} for a in plan.pop("answers", [])]
plan["needs_confirm"] = [a["type"] for a in plan["actions"] if a["type"] in pc.CONFIRM]
if dropped:
plan["dropped"] = dropped
if plan["answered"] and not plan["actions"]:
plan["reply"] = plan["answered"][0]["text"]
return plan
def prompt_answer(self, job_id, topic, ctx):
"""Read-only answers, from the job's own data (never a model's opinion)."""
path = self.job_dir(job_id)
wo = json.loads((path / "work_order.json").read_text()) if (path / "work_order.json").is_file() else None
objs = ctx["objects"]
if topic == "objects":
counts = {}
for o in objs:
counts[o["label"]] = counts.get(o["label"], 0) + 1
return ("Detected on this wall: " + ", ".join(f"{n} x {k.lower()}" for k, n in sorted(counts.items())) + ".") if counts else "No windows, doors or fixtures were detected."
if topic in ("paint", "area"):
src = ((wo or {}).get("scale") or {}).get("source", "unknown")
assumed = "assumed" in str(src)
if (path / "twin.npz").is_file():
import twin_model
_, twin = self._twin_load(job_id)
ev = self.evidence(job_id)
unc = twin_model.size_uncertainty(twin, ev.get("typical_error_pct"), ev.get("high_error_pct"))
q = twin_model.quantities(twin, uncertainty=unc)
if topic == "area":
return (f"The twin's wall is {twin.width_mm / 1000:.2f} x {twin.height_mm / 1000:.2f} m ({twin.width_mm * twin.height_mm / 1e6:.1f} m2 gross, "
f"{q['paintable_wall_m2']:.1f} m2 paintable wall), size from {twin.accuracy()['dimension_source']}"
+ (f", about +/-{unc['area_pm_pct']:.0f}% in area" if unc.get("known") else "") + ".")
return (f"About {q['litres']:g} L for {q['coats']} coats of the {q['to_paint_m2']:.1f} m2 that is unpainted and seen (coverage {q['coverage_m2_per_l']:g} m2/L, "
f"{int(q['waste'] * 100)}% waste); {q['unobserved_m2']:.1f} m2 has not been seen. " + (q.get("warning", "") or "Use the tin's own coverage figure."))
if wo and wo.get("summary", {}).get("paint_area_ft2") is not None:
a = wo["summary"]["paint_area_ft2"]
return (f"The plan covers about {a:.1f} sq ft ({a * 0.0929:.1f} m2): roughly {a * 0.0929 * 0.1 * 2:.1f} L for two coats at 10 m2/L. "
+ ("The scale is an ASSUMED 8 ft ceiling, so this is not trustworthy: give the real wall height. " if assumed else "") + "Build the twin for a better estimate.")
return "There is no plan yet for this job, so there is no area to report. Press Re-plan."
if topic == "accuracy":
ev = self.evidence(job_id)
return f"Evidence score {ev['score']}% ({ev['label']}): {ev['expected_error']}. That is how much evidence there is, not a probability of being right."
if topic == "drone":
try:
st = self.dronetwin_status(job_id)
s = st["status"]
return f"Drone twin: {st['sorties']} sortie(s) logged, verdict {s.get('verdict')}, about {round(100 * s.get('believed_painted', 0))}% believed painted. (Set up a facade first if this is zero.)"
except Exception:
return "No drone twin is set up for this job yet (it needs a home position and the facade's end points)."
return "I can't answer that yet."
def prompt_apply(self, job_id, request):
"""Run the proposed actions. Re-validates everything: only whitelisted verbs with sane values run; each result says what happened."""
import prompt_cmd as pc
path = self.job_dir(job_id)
ctx = self.prompt_context(job_id)
actions, dropped = pc.validate(request.get("actions"), ctx)
results, client = [], []
for a in actions:
t = a["type"]
try:
if t == "zone":
cur = self.get_zones(job_id)
zones = cur["zones"] + [{"kind": a["kind"], "shape": a["shape"], "pts": a["pts"], "label": a["label"] or "from instruction", "text": "added by a typed instruction", "visible": True}]
self.set_zones(job_id, {"zones": zones, "info": cur.get("info")})
results.append({"type": t, "ok": True, "detail": f"added zone '{a['label']}'; Re-plan to use it"})
elif t == "clear_zones":
cur = self.get_zones(job_id)
self.set_zones(job_id, {"zones": [], "info": cur.get("info")})
results.append({"type": t, "ok": True, "detail": f"removed {len(cur['zones'])} zone(s); Re-plan to use that"})
elif t == "set_height":
out = self.set_scale_from_drawing(job_id, {"length_mm": a["feet"] * 304.8, "axis": "height", "source": "operator", "label": "typed instruction"})
results.append({"type": t, "ok": True, "detail": f"wall height {a['feet']:g} ft recorded as your measurement; evidence now {out['evidence']['score']}%"})
elif t == "twin":
import twin_pipeline
rep = twin_pipeline.run(self, job_id, {})
results.append({"type": t, "ok": rep["ok"], "detail": next((s["detail"] for s in rep["stages"] if s["name"] == "twin"), "")})
elif t == "drawing":
d = self.generate_drawing(job_id, {})
results.append({"type": t, "ok": True, "detail": f"drew {d['openings']} opening(s) at {d['wall_mm'][0] / 1000:.2f} x {d['wall_mm'][1] / 1000:.2f} m (draft); see Drawings", "drawing": d["id"]})
elif t in ("layer", "replan"):
client.append(a) # the browser does these with its own controls
results.append({"type": t, "ok": True, "detail": a.get("why") or t})
except PermissionError as exc:
results.append({"type": t, "ok": False, "detail": str(exc)})
except (ValueError, KeyError) as exc:
results.append({"type": t, "ok": False, "detail": str(exc)})
return {"results": results, "client_actions": client, "dropped": dropped}
def twin_cell(self, job_id, x_mm, y_mm):
"""What the twin holds at one point of the wall (millimetres from its top-left): the surface, paint, how well and how recently seen, relief, and any opening there."""
import twin as tn
import twin_model
path, twin = self._twin_load(job_id)
x, y = float(x_mm), float(y_mm)
if not (0 <= x < twin.width_mm and 0 <= y < twin.height_mm):
raise ValueError("that point is outside the wall")
c, r = int(x // twin.cell_mm), int(y // twin.cell_mm)
names = {tn.UNKNOWN: "not observed", **{v: k for k, v in {"unpainted wall": tn.UNPAINTED, "painted wall": tn.PAINTED, "uncertain wall": tn.UNCERTAIN, "skirting": tn.SKIRTING,
"switch / outlet": tn.SWITCH, "air conditioner": tn.AC, "door": tn.DOOR, "window": tn.WINDOW, "obstacle": tn.OBSTACLE}.items()}}
now_s = self._twin_clock(path)
seen = float(twin.weight[r, c]) > 0.05 or int(twin.surface[r, c]) != tn.UNKNOWN
inside = [o["id"] for o in twin_model.openings(twin) if o["x_mm"] <= x < o["x_mm"] + o["w_mm"] and o["y_mm"] <= y < o["y_mm"] + o["h_mm"]]
out = {"x_mm": round(x), "y_mm": round(y), "surface": names.get(int(twin.surface[r, c]), "other"), "observed": bool(seen),
"painted_fraction": round(float(twin.painted[r, c]), 2) if twin.weight[r, c] > 0.05 else None, "evidence_weight": round(float(twin.weight[r, c]), 2),
"age_s": round(now_s - float(twin.last_seen[r, c]), 1) if twin.last_seen[r, c] >= 0 else None, "planned": bool(twin.effective_plan()[r, c]),
"damage": bool(twin.damage[r, c]), "relief_mm": round(float(twin.relief[r, c]), 1) if twin.relief_w[r, c] > 0 else None, "opening": inside[0] if inside else None}
return out
def twin_history(self, job_id):
"""The snapshots a pipeline run saved: how much paint was seen on the wall at each, and what changed between consecutive ones."""
import twin_model
from twin import WALL_CLASSES
path = self.job_dir(job_id)
snaps = sorted((path / "twin_snapshots").glob("*.npz")) if (path / "twin_snapshots").is_dir() else []
out, prev = [], None
for f in snaps:
z = np.load(f, allow_pickle=False)
cur = {"surface": z["surface"], "painted": z["painted"], "damage": z["damage"], "dims": tuple(z["dims"])}
ft2 = (cur["dims"][2] / 304.8) ** 2
wall = np.isin(cur["surface"], WALL_CLASSES)
row = {"snapshot": f.stem, "painted_ft2": round(float((cur["painted"] & wall).sum()) * ft2, 1), "paintable_ft2": round(float(wall.sum()) * ft2, 1),
"observed_share": round(float((cur["surface"] != 255).mean()), 3)}
if prev is not None:
d = twin_model.diff(prev, cur)
row["change"] = d if d.get("comparable") else {"comparable": False, "reason": d.get("reason")}
out.append(row)
prev = cur
return {"snapshots": out}
def room_twin(self, request):
"""Combine the twins of several jobs (walls in order around a room) into one room summary; stateless, and nothing is stored."""
import room_twin
ids = request.get("jobs") or []
if not isinstance(ids, list) or not 1 <= len(ids) <= 12:
raise ValueError("give between 1 and 12 job ids, in order around the room")
walls = []
for jid in ids:
path, twin = self._twin_load(str(jid))
info = json.loads((path / "job.json").read_text())
walls.append((info.get("wall_id") or info.get("filename") or str(jid), twin))
room = room_twin.build(walls, coats=int(request.get("coats") or 2))
if request.get("obj"):
room["obj"] = room_twin.to_obj(room, str(request.get("name") or "room"))
return room
def twin_export(self, job_id, name):
"""The twin as data for other tools: its openings in wall millimetres (JSON / CSV) or a heightfield mesh (OBJ, metres)."""
import twin_model
path, twin = self._twin_load(job_id)
if name == "model.obj":
return "text/plain; charset=utf-8", twin_model.export_obj(twin)
if name == "defects.json":
return "application/json", {"origin": "top-left of the wall, millimetres", "defects": twin_model.defects(twin)}
if name == "quantities.json":
return "application/json", twin_model.quantities(twin)
rows = twin_model.openings(twin)
if name == "openings.csv":
return "text/csv; charset=utf-8", twin_model.openings_csv(rows)
return "application/json", {"wall_mm": [round(twin.width_mm), round(twin.height_mm)], "origin": "top-left of the wall, x right, y down, millimetres",
"dimension_source": twin.accuracy()["dimension_source"], "openings": rows}
def twin_state(self, job_id):
path, twin = self._twin_load(job_id)
s = twin.summary(self._twin_clock(path))
s["history"] = twin.history[-30:]
s["layers"] = list(self.TWIN_LAYERS)
s["live_linked"] = bool(getattr(self, "live", None) and self.live.get("twin_job") == job_id)
s["sources"] = json.loads((path / "twin_meta.json").read_text()).get("sources", {})
try:
ev = self.evidence(job_id)
s["model_accuracy"] = {k: ev[k] for k in ("score", "label", "expected_error", "typical_error_pct", "high_error_pct", "next", "cap", "conflict", "disclaimer")}
except Exception: # the accuracy block is advisory: a failure there must not hide the twin
s["model_accuracy"] = None
return s
def twin_layer(self, job_id, layer):
if layer not in self.TWIN_LAYERS:
raise ValueError(f"unknown layer {layer!r}: {', '.join(self.TWIN_LAYERS)}")
path, twin = self._twin_load(job_id)
return twin.to_png(layer, self._twin_clock(path))
def twin_observe_live(self, job_id, rgb, masks, summary):
"""Called by a live session for each locked frame: register the frame to the job's photo and fold it into the twin."""
import cv2
import evidence_inputs as ei
path, twin = self._twin_load(job_id)
gray_main, labels_main = self._main_arrays(path)
labels = masks.get("mask")
if labels is None:
return
gray = cv2.cvtColor(np.asarray(rgb), cv2.COLOR_RGB2GRAY)
H, inl, why = ei.register_homography(gray_main, labels_main, gray, np.asarray(labels, np.uint8))
if H is None:
return
dmg = masks.get("hazards")
twin.observe_view(np.asarray(labels, np.uint8), H, self._twin_clock(path), "live frame", weight=0.8)
(path / "twin.npz").write_bytes(twin.to_bytes())
def auto_match_drawing(self, job_id, drawing_id, request):
"""Place the job's photo on an uploaded drawing from the layout of its windows and doors (drawing_match.py), instead of by hand.
Needs the drawing calibrated first (its two clicked points + real length give mm per pixel, so its openings read in mm) and the
job's saved class mask. The result is stored on the drawing (``auto_match``) so the operator can review it. The scale is only
written to the job when the request says ``apply`` AND the match is usable (two or more openings agree and no other placement fits
as well); an ambiguous or one-opening match is reported and never applied."""
import cv2
import drawing_match as dm
path, items = self._drawings(job_id)
item = next((d for d in items if d["id"] == drawing_id), None)
if item is None:
raise KeyError(drawing_id)
cal = item.get("calibration")
if not cal:
raise ValueError("calibrate the drawing first (two clicked points and a real length): its scale is needed to read its openings")
job = self.job_dir(job_id)
classes = job / "classes.png"
if not classes.is_file():
raise ValueError("this job has no saved class mask; re-run the photo through the current model")
labels = np.asarray(Image.open(classes))
gray = cv2.imread(str(path / f"{drawing_id}.png"), cv2.IMREAD_GRAYSCALE)
try:
drawing = dm.drawing_from_image(gray, float(cal["mm_per_px"]))
except ValueError as exc:
raise ValueError(f"could not read openings from this drawing: {exc}") from exc
result = dm.match(dm.openings_from_mask(labels), drawing, labels.shape)
applied = False
if request.get("apply"):
if json.loads((job / "job.json").read_text()).get("approved"):
raise PermissionError("job is approved and frozen; start a new job to change the scale")
if not (result.get("matched") and result.get("usable_for_scale")):
raise ValueError("this match is not usable for scale (" + str(result.get("note") or result.get("reason") or "low confidence") + ")")
saved = {"mm_per_px": result["mm_per_px_at_centre"], "source": f"drawing auto-match ({result['confidence']} confidence, "
f"{len(result['matched_openings'])} openings)", "axis": "match", "length_mm": None, "at": now()}
(job / "scale.json").write_text(json.dumps(saved, indent=1))
self._record_anchor(job, "drawing", result["mm_per_px_at_centre"], f"drawing auto-match ({result['confidence']})")
applied = True
item["auto_match"] = {k: result.get(k) for k in ("matched", "confidence", "ambiguous", "usable_for_scale", "mm_per_px_at_centre",
"matched_openings", "in_photo_not_in_drawing", "in_drawing_not_in_photo",
"candidates", "note", "reason", "photo_footprint_mm", "wall_mm")} | {
"applied": applied, "at": now()}
(path / "index.json").write_text(json.dumps(items, indent=1))
return {"match": item["auto_match"], "applied": applied}
def list_jobs(self):
jobs = []
for path in sorted(self.jobs.iterdir(), key=lambda p: p.stat().st_mtime, reverse=True):
if JOB_ID.match(path.name) and (path / "job.json").is_file():
jobs.append(json.loads((path / "job.json").read_text()))
return jobs[:200]
def create_job(self, data, filename="", wall_id=""):
if len(data) > MAX_UPLOAD:
raise ValueError(f"image larger than {MAX_UPLOAD // (1024 * 1024)} MB")
try:
image = Image.open(io.BytesIO(data))
image.load()
except Exception as exc:
raise ValueError(f"not a readable image: {exc}") from exc
from PIL import ImageOps
image = ImageOps.exif_transpose(image).convert("RGB")
if max(image.size) > MAX_SIDE:
image.thumbnail((MAX_SIDE, MAX_SIDE), Image.Resampling.LANCZOS)
rgb = np.asarray(image)
job_id = secrets.token_hex(6)
path = self.jobs / job_id
path.mkdir()
image.save(path / "photo.jpg", quality=92)
with self.lock:
started = time.time()
summary, masks = self.analyze_fn(rgb)
seconds = round(time.time() - started, 2)
if masks.get("mask") is not None and self.defaults.get("window_gate", False):
masks["mask"], summary["window_lookalike_gate"] = window_gate(rgb, masks["mask"])
(path / "result.json").write_text(json.dumps(summary, indent=2) + "\n")
save_mask(path / "paintable.png", masks["safe_paintable"])
save_mask(path / "keepout.png", masks["hazards"])
save_mask(path / "touchup.png", masks["touch_up"])
save_mask(path / "wall.png", masks["wall"])
if masks.get("mask") is not None:
Image.fromarray(np.asarray(masks["mask"], np.uint8)).save(path / "classes.png")
layer_png(masks["uncertain"], LAYER_COLOURS["uncertain"]).save(path / "layer_uncertain.png")
edge_png(masks["uncertain"], LAYER_COLOURS["uncertain"]).save(path / "layer_uncertain_edge.png")
job = {"id": job_id, "created": now(), "filename": Path(filename).name[:120], "wall_id": wall_id[:80],
"width": image.size[0], "height": image.size[1], "photo_sha256": sha256_bytes(
(path / "photo.jpg").read_bytes()), "analysis_seconds": seconds, "model": self.model_info,
"status": None, "approved": None}
(path / "job.json").write_text(json.dumps(job, indent=2) + "\n")
self.plan(job_id, {})
return self.read_job(job_id)
def plan(self, job_id, request):
path = self.job_dir(job_id)
job = json.loads((path / "job.json").read_text())
if job.get("approved"):
raise PermissionError("job is approved and frozen; start a new job to change the plan")
prediction = json.loads((path / "result.json").read_text())
paintable, keepout = load_mask(path / "paintable.png"), load_mask(path / "keepout.png")
touchup, wall = load_mask(path / "touchup.png"), load_mask(path / "wall.png")
# edits_png absent: keep the saved edits; "" clears them; a data URL replaces them.
if "edits_png" in request:
edits = decode_edits(request["edits_png"], paintable.shape)
if edits is not None and edits.any():
Image.fromarray(edits).save(path / "edits.png")
elif (path / "edits.png").is_file():
(path / "edits.png").unlink()
edits = np.asarray(Image.open(path / "edits.png")) if (path / "edits.png").is_file() else None
edit_stats = None
operator_zones = [z for z in self.get_zones(job_id)["zones"] if z.get("visible", True)]
effective = edits
if any(z["kind"] != "note" for z in operator_zones):
zmap = self.zone_edits(operator_zones, paintable.shape)
effective = zmap if edits is None else np.where(zmap > 0, zmap, edits).astype(edits.dtype)
if effective is not None:
paintable, keepout, edit_stats = apply_edits(paintable, keepout, effective, wall=wall)
if edits is not None: # the brush layer only: zones are drawn from zones.json, not baked into it
rgba = np.zeros(edits.shape + (4,), np.uint8)
rgba[edits == EDIT_ADD] = (0, 255, 0, 255)
rgba[edits == EDIT_EXCLUDE] = (255, 0, 0, 255)
Image.fromarray(rgba).save(path / "layer_edits.png")
if edits is None and (path / "layer_edits.png").is_file():
(path / "layer_edits.png").unlink()
operator = (request.get("operator") or "").strip()[:80]
confirmations = {}
for name in request.get("confirmations") or []:
if name not in CONFIRMATIONS:
raise ValueError(f"unknown confirmation {name!r}")
if not operator:
raise ValueError("enter the operator's name to sign a confirmation")
confirmations[name] = {"by": operator, "at": now()}
depth = request.get("depth_report")
if depth is None and (path / "depth_report.json").is_file():
depth = json.loads((path / "depth_report.json").read_text()) # a depth scan uploaded for this job (evidence.py)
saved = json.loads((path / "scale.json").read_text()) if (path / "scale.json").is_file() else {}
mm_per_px = request.get("mm_per_px") or saved.get("mm_per_px") or self.defaults.get("mm_per_px")
source = "operator" if request.get("mm_per_px") else (saved.get("source") if saved.get("mm_per_px")
else ("default" if mm_per_px else None))
if not mm_per_px:
mm_per_px, source = scale_from_prediction(prediction)
tools_mm = {**{"roller": 230.0, "trim": 60.0, "prep": 125.0, "margin": 40.0}, **(request.get("tools_mm") or {})}
order, _ = build_work_order(prediction, paintable, keepout, touchup, depth=depth,
require_depth=self.defaults.get("require_depth", True),
mm_per_px=float(mm_per_px) if mm_per_px else None, scale_source=source,
tools_mm=tools_mm, confirmations=confirmations, edit_stats=edit_stats)
order["job_id"], order["planned_at"], order["model"] = job_id, now(), self.model_info
zdata = self.get_zones(job_id)
order["operator_zones"] = [{"kind": z["kind"], "shape": z["shape"], "label": z["label"], "text": z["text"],
"points_px": z["pts"], "applied": z["kind"] != "note" and z.get("visible", True), "correction": z["kind"] in CORRECTION_KINDS}
for z in zdata["zones"]]
order["job_info"] = zdata["info"]
(path / "work_order.json").write_text(json.dumps(order, indent=2) + "\n")
layer_png(paintable & ~keepout, LAYER_COLOURS["paint"], 120).save(path / "layer_paint.png")
layer_png(keepout & wall, LAYER_COLOURS["keepout"], 120).save(path / "layer_keepout.png")
edge_png(paintable & ~keepout, LAYER_COLOURS["paint"]).save(path / "layer_paint_edge.png")
edge_png(keepout & wall, LAYER_COLOURS["keepout"]).save(path / "layer_keepout_edge.png")
# Touch-up patches only matter on a signed-off wall; otherwise the coat covers them.
signed_off = (prediction.get("completion") or {}).get("status") == "complete"
layer_png(touchup & ~keepout & signed_off, LAYER_COLOURS["touchup"], 170).save(path / "layer_touchup.png")
edge_png(touchup & ~keepout & signed_off, LAYER_COLOURS["touchup"]).save(path / "layer_touchup_edge.png")
job.update(status=order["status"], summary=order.get("summary"), updated=now())
(path / "job.json").write_text(json.dumps(job, indent=2) + "\n")
return self.read_job(job_id)
def demo(self, job_id):
"""Tape + paint plan as drawable geometry (see demo_plan.py) for the phone demo page."""
from demo_plan import build_demo_plan
path = self.job_dir(job_id)
if not (path / "classes.png").is_file():
raise ValueError("this job predates the demo plan: upload the photo again")
prediction = json.loads((path / "result.json").read_text())
classes = self.corrected_classes(np.asarray(Image.open(path / "classes.png")), self.get_zones(job_id)["zones"])
from scipy import ndimage
from demo_plan import CLASS_IDS, WALL_IDS
protect = np.isin(classes, list(CLASS_IDS.values()))
paintable = np.isin(classes, WALL_IDS) & ~ndimage.binary_dilation(protect, iterations=3)
layer_png(paintable, (255, 255, 255), 255).save(path / "demo_paint.png")
from demo_plan import regions_rgba
Image.fromarray(regions_rgba(classes)[0]).save(path / "demo_regions.png")
gray = np.asarray(Image.open(path / "photo.jpg").convert("L"))
mm_per_px, source = scale_from_prediction(prediction)
plan = build_demo_plan(classes, paintable, mm_per_px, source, gray)
plan["job_id"], plan["status"] = job_id, json.loads((path / "job.json").read_text()).get("status")
return plan
def create_video(self, data, filename=""):
"""Save an upload and analyse keyframes (video_demo.py). Synchronous: the model is the slow part."""
from video_demo import process_video
ext = Path(filename).suffix.lstrip(".").lower()
if ext not in VIDEO_EXT:
raise ValueError(f"video must be one of: {', '.join(sorted(VIDEO_EXT))}")
if not data or len(data) > MAX_VIDEO:
raise ValueError(f"video must be between 1 byte and {MAX_VIDEO // (1024 * 1024)} MB")
vid = "v" + secrets.token_hex(6)
path = self.jobs / vid
path.mkdir()
(path / f"video.{ext}").write_bytes(data)
try:
with self.lock:
manifest = process_video(path / f"video.{ext}", self.analyze_fn, path)
except ValueError:
import shutil
shutil.rmtree(path, ignore_errors=True)
raise
manifest.update(id=vid, video=f"video.{ext}")
(path / "manifest.json").write_text(json.dumps(manifest, indent=1))
return manifest
def approve(self, job_id, request):
from run_registry import append
operator = (request.get("operator") or "").strip()[:80]
if not operator:
raise ValueError("enter the operator's name to approve")
with self.lock:
path = self.job_dir(job_id)
job = json.loads((path / "job.json").read_text())
if job.get("approved"):
raise PermissionError("already approved")
order_bytes = (path / "work_order.json").read_bytes()
order = json.loads(order_bytes)
if order["status"] == "blocked":
raise PermissionError("the safety gate blocks this order: " + "; ".join(order["gate"]["reasons"]))
hashes = {name: sha256_bytes((path / name).read_bytes())
for name in ("photo.jpg", "result.json", "paintable.png", "keepout.png", "edits.png")
if (path / name).is_file()}
approval = {"job_id": job_id, "operator": operator, "note": (request.get("note") or "")[:500],
"approved_at": now(), "work_order_sha256": sha256_bytes(order_bytes), "inputs_sha256": hashes,
"status": order["status"], "confirmations": order.get("confirmations"),
"operator_edits": order.get("operator_edits"), "model": self.model_info}
record = append(self.jobs / "registry.jsonl", "work_order_approval",
checkpoint_sha256=(self.model_info or {}).get("sha256"), run_id=job_id,
config={"work_order_sha256": approval["work_order_sha256"]},
extra={"operator": operator, "work_order_sha256": approval["work_order_sha256"]})
approval["registry_hash"] = record["hash"]
(path / "approval.json").write_text(json.dumps(approval, indent=2) + "\n")
job.update(approved={"by": operator, "at": approval["approved_at"], "registry_hash": record["hash"]})
(path / "job.json").write_text(json.dumps(job, indent=2) + "\n")
return self.read_job(job_id)
def make_handler(console):
class Handler(BaseHTTPRequestHandler):
server_version = f"PaintingVisionConsole/{VERSION}"
server_console = console
def log_message(self, fmt, *args): # quieter, one line per request
print(f"{self.address_string()} {fmt % args}", flush=True)
def send(self, status, body, content_type="application/json", extra=None):
if isinstance(body, (dict, list)):
body = json.dumps(body).encode()
elif isinstance(body, str):
body = body.encode()
self.send_response(status)
self.send_header("Content-Type", content_type)
self.send_header("Content-Length", str(len(body)))
self.send_header("Cache-Control", "no-store")
self.send_header("X-Content-Type-Options", "nosniff")
self.send_header("Content-Security-Policy", "default-src 'self'; img-src 'self' data: blob:; "
"style-src 'self'; script-src 'self'; media-src 'self'")
for key, value in (extra or {}).items():
self.send_header(key, value)
self.end_headers()
self.wfile.write(body)
def send_file(self, path, name):
"""Serve a job file with Range support (Safari will not play a video without it)."""
ext = name.rsplit(".", 1)[-1]
ctype = VIDEO_EXT.get(ext) or {"json": "application/json", "png": "image/png", "jpg": "image/jpeg"}[ext]
size = path.stat().st_size
rng = re.match(r"^bytes=(\d*)-(\d*)$", self.headers.get("Range") or "")
start, end, status = 0, size - 1, 200
if rng and (rng.group(1) or rng.group(2)):
if rng.group(1):
start = int(rng.group(1))
end = int(rng.group(2)) if rng.group(2) else size - 1
else:
start = max(0, size - int(rng.group(2)))
end = min(end, size - 1)
if start > end:
self.send_response(416)
self.send_header("Content-Range", f"bytes */{size}")
self.end_headers()
return
status = 206
with open(path, "rb") as f:
f.seek(start)
body = f.read(end - start + 1)
self.send_response(status)
self.send_header("Content-Type", ctype)
self.send_header("Content-Length", str(len(body)))
self.send_header("Accept-Ranges", "bytes")
self.send_header("Cache-Control", "no-store")
self.send_header("X-Content-Type-Options", "nosniff")
if status == 206:
self.send_header("Content-Range", f"bytes {start}-{end}/{size}")
self.end_headers()
self.wfile.write(body)
def error(self, status, message):
self.send(status, {"error": message})
def body(self, limit=MAX_UPLOAD):
length = int(self.headers.get("Content-Length") or 0)
if length > limit:
raise ValueError("request too large")
return self.rfile.read(length) if length else b""
def do_GET(self):
url = urlparse(self.path)
parts = [p for p in url.path.split("/") if p]
try:
if not parts:
return self.send(200, (STATIC / "index.html").read_bytes(), STATIC_FILES["index.html"])
if parts == ["api", "assist", "status"]:
import ai_assist
return self.send(200, ai_assist.status())
if parts == ["api", "health"]:
return self.send(200, {"ok": True, "version": VERSION, "model": console.model_info,
"confirmations": {k: v[1] for k, v in CONFIRMATIONS.items()}})
if parts == ["api", "jobs"]:
return self.send(200, console.list_jobs())
if len(parts) == 3 and parts[:2] == ["api", "jobs"]:
return self.send(200, console.read_job(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "zones":
return self.send(200, console.get_zones(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "dronetwin":
return self.send(200, console.dronetwin_status(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "dronetwinnext":
return self.send(200, console.dronetwin_next(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "dronetwinwhatif":
q = parse_qs(url.query)
return self.send(200, console.dronetwin_whatif(parts[2], float(q.get("positioning", [0.02])[0]),
float(q.get("wind", [2.0])[0]), int(q.get("trials", [3])[0])))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "floor":
return self.send(200, console.floor(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "objects":
return self.send(200, console.objects(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings":
return self.send(200, console.list_drawings(parts[2]))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings" and parts[4].rsplit(".", 1)[-1] in ("svg", "dxf", "json"):
did, ext = parts[4].rsplit(".", 1)
mime = {"svg": "image/svg+xml", "dxf": "application/dxf", "json": "application/json"}[ext]
f = console.drawing_export(parts[2], did, ext)
return self.send(200, f.read_bytes(), mime, {"Content-Disposition": f'attachment; filename="{parts[2]}_{did}.{ext}"'}
if "download" in parse_qs(url.query) else None)
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings":
return self.send(200, console.drawing_file(parts[2], parts[4].removesuffix(".png")).read_bytes(), "image/png")
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "twin":
try:
return self.send(200, console.twin_state(parts[2]) | {"built": True})
except KeyError as exc:
if "no twin" not in str(exc):
raise
return self.send(200, {"built": False}) # not an error: most jobs simply have no twin yet
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4] == "pipeline":
import twin_pipeline
return self.send(200, twin_pipeline.last(console, parts[2]))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4] == "history":
return self.send(200, console.twin_history(parts[2]))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4] == "cell":
q = parse_qs(url.query)
return self.send(200, console.twin_cell(parts[2], q.get("x", ["nan"])[0], q.get("y", ["nan"])[0]))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4] in ("openings.json", "openings.csv", "model.obj", "defects.json", "quantities.json"):
kind, body = console.twin_export(parts[2], parts[4])
return self.send(200, body, kind, {"Content-Disposition": f'attachment; filename="{parts[2]}_twin_{parts[4]}"'} if "download" in parse_qs(url.query) else None)
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4].endswith(".png"):
return self.send(200, console.twin_layer(parts[2], parts[4][:-4]), "image/png")
if parts == ["api", "videos"]:
return self.send(200, console.list_videos())
if parts == ["api", "live"]:
return self.send(200, console.live_state())
if parts == ["api", "live", "frame.jpg"]:
return self.send(200, console.live_frame(), "image/jpeg")
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "evidence":
return self.send(200, console.evidence(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "demo":
return self.send(200, console.demo(parts[2]))
if len(parts) == 4 and parts[:2] == ["api", "videos"] and VIDEO_ID.match(parts[2]) \
and VIDEO_FILE.match(parts[3]):
path = self.server_console.jobs / parts[2] / parts[3]
if not path.is_file():
return self.error(404, "not found")
return self.send_file(path, parts[3])
if parts == ["demo"]:
return self.send(200, (STATIC / "demo.html").read_bytes(), STATIC_FILES["demo.html"])
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] in JOB_FILES:
path = console.job_dir(parts[2]) / parts[3]
if not path.is_file() and parts[3].endswith("_edge.png"):
# a job made before outlines existed: derive the ring from its fill layer's own pixels
name = parts[3][len("layer_"):-len("_edge.png")]
fill = path.parent / f"layer_{name}.png"
if name in LAYER_COLOURS and fill.is_file():
edge_png(np.asarray(Image.open(fill).convert("RGBA"))[..., 3] > 0, LAYER_COLOURS[name]).save(path)
if not path.is_file():
return self.error(404, "not found")
extra = ({"Content-Disposition": f'attachment; filename="{parts[2]}_{parts[3]}"'}
if "download" in parse_qs(url.query) else None)
return self.send(200, path.read_bytes(), JOB_FILES[parts[3]], extra)
if len(parts) == 1 and parts[0] in STATIC_FILES:
return self.send(200, (STATIC / parts[0]).read_bytes(), STATIC_FILES[parts[0]])
return self.error(404, "not found")
except KeyError:
return self.error(404, "no such job")
except (ValueError, json.JSONDecodeError) as exc: # a bad parameter in a GET is the caller's error, not a dropped connection
return self.error(400, str(exc))
except Exception as exc:
import traceback
traceback.print_exc()
return self.error(500, f"{type(exc).__name__}: {exc}")
def do_POST(self):
url = urlparse(self.path)
parts = [p for p in url.path.split("/") if p]
try:
if parts == ["api", "videos"]:
query = parse_qs(url.query)
return self.send(201, console.create_video(self.body(MAX_VIDEO), query.get("filename", [""])[0]))
if parts == ["api", "jobs"]:
query = parse_qs(url.query)
job = console.create_job(self.body(), query.get("filename", [""])[0], query.get("wall_id", [""])[0])
return self.send(201, job)
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "laser":
return self.send(201, console.add_laser(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "scan":
q = parse_qs(url.query)
return self.send(201, console.add_scan(parts[2], self.body(MAX_SCAN), q.get("filename", [""])[0], q.get("up", ["z"])[0], q.get("flip", ["0"])[0] in ("1", "true")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "twin":
return self.send(201, console.twin_build(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4] == "pipeline":
import twin_pipeline
return self.send(200, twin_pipeline.run(console, parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "twin" and parts[4] == "update":
return self.send(200, console.twin_update(parts[2]))
if parts == ["api", "live"]:
return self.send(201, console.live_start(json.loads(self.body() or b"{}")))
if parts == ["api", "live", "arm"]:
return self.send(200, console.live_arm(json.loads(self.body() or b"{}")))
if parts == ["api", "live", "stop"]:
return self.send(200, console.live_stop())
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "views":
return self.send(201, console.add_view(parts[2], self.body(MAX_VIDEO), parse_qs(url.query).get("filename", [""])[0]))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "depth":
q = parse_qs(url.query)
num = lambda k: (float(q[k][0]) if q.get(k) and q[k][0] != "" else None) # noqa: E731
return self.send(201, console.add_depth(parts[2], self.body(MAX_DEPTH), q.get("filename", [""])[0], fx=num("fx"), hfov_deg=num("hfov_deg")))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings" and parts[4] == "generate":
return self.send(201, console.generate_drawing(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings":
query = parse_qs(url.query)
return self.send(201, console.add_drawing(parts[2], self.body(MAX_DRAWING), query.get("filename", [""])[0]))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings" and DRAWING_ID.match(parts[4]):
return self.send(200, console.update_drawing(parts[2], parts[4], json.loads(self.body() or b"{}")))
if len(parts) == 6 and parts[:2] == ["api", "jobs"] and parts[3] == "drawings" and DRAWING_ID.match(parts[4]) \
and parts[5] == "match":
return self.send(200, console.auto_match_drawing(parts[2], parts[4], json.loads(self.body() or b"{}")))
if parts == ["api", "assist"]:
# Optional AI helper: advisory text only, and only with an API key in the server's environment (ai_assist.py).
import ai_assist
request = json.loads(self.body() or b"{}")
if not isinstance(request, dict):
raise ValueError("expected a JSON object")
return self.send(200, ai_assist.ask(request.get("context"), request.get("question", "")))
if parts == ["api", "room"]:
return self.send(200, console.room_twin(json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "prompt":
return self.send(200, console.prompt(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 5 and parts[:2] == ["api", "jobs"] and parts[3] == "prompt" and parts[4] == "apply":
return self.send(200, console.prompt_apply(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "zones":
return self.send(200, console.set_zones(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "dronetwin":
return self.send(200, console.dronetwin_status(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "dronetwinobserve":
return self.send(200, console.dronetwin_observe(parts[2], self.body()))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "dronetwinlog":
sortie = parse_qs(url.query).get("sortie", [None])[0]
return self.send(200, console.dronetwin_log(parts[2], self.body(MAX_DRAWING).decode("utf-8", "replace"), sortie))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] == "scale":
return self.send(200, console.set_scale_from_drawing(parts[2], json.loads(self.body() or b"{}")))
if len(parts) == 4 and parts[:2] == ["api", "jobs"] and parts[3] in ("plan", "approve"):
request = json.loads(self.body() or b"{}")
if not isinstance(request, dict):
raise ValueError("expected a JSON object")
action = console.plan if parts[3] == "plan" else console.approve
return self.send(200, action(parts[2], request))
return self.error(404, "not found")
except KeyError:
return self.error(404, "no such job")
except PermissionError as exc:
return self.error(409, str(exc))
except (ValueError, json.JSONDecodeError) as exc:
return self.error(400, str(exc))
except Exception as exc: # keep the console up; the operator sees the error
import traceback
traceback.print_exc()
return self.error(500, f"{type(exc).__name__}: {exc}")
return Handler
def model_analyzer(checkpoint_path, device_name=None):
"""Load the checkpoint once; return ``(analyze_fn, model_info)``."""
import cv2
import torch
from hub import resolve_checkpoint
from predict import analyze, build_parser
from train import load_checkpoint_model
path = Path(resolve_checkpoint(str(checkpoint_path)))
device = torch.device(device_name or ("cuda" if torch.cuda.is_available() else "cpu"))
model, checkpoint = load_checkpoint_model(path, device)
args = build_parser().parse_args(["unused.jpg"])
args.depth_map, args.focal_px = None, None
warnings = []
if checkpoint.get("completion_threshold") is None:
warnings.append("sign-off is disabled: this checkpoint has no calibrated completion threshold")
if not checkpoint.get("condition_operating_point"):
warnings.append("damage flags use uncalibrated defaults (run calibrate_condition.py --write)")
if "temperature" not in checkpoint:
warnings.append("confidences are not temperature-calibrated (run train.py --calibrate-checkpoint)")
info = {"checkpoint": path.name, "sha256": hashlib.sha256(path.read_bytes()).hexdigest(),
"arch": checkpoint.get("arch"), "device": str(device),
"completion_threshold": checkpoint.get("completion_threshold"),
"condition_operating_point": checkpoint.get("condition_operating_point"), "warnings": warnings}
def analyze_fn(rgb):
return analyze(cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR), model, checkpoint, device, args)
return analyze_fn, info
def main():
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("--checkpoint", default="artifacts/best.pt", help="local .pt or hf://owner/repo")
ap.add_argument("--jobs-dir", type=Path, default=Path("jobs"))
ap.add_argument("--host", default="127.0.0.1", help="0.0.0.0 exposes the console to the network: only on a "
"trusted site network")
ap.add_argument("--port", type=int, default=8765)
ap.add_argument("--device")
ap.add_argument("--mm-per-px", type=float, help="default scale when a job gives none")
ap.add_argument("--no-window-gate", action="store_true",
help="keep every window the wall model finds (skip the picture/mirror/screen look-alike check)")
ap.add_argument("--no-require-depth", action="store_true",
help="do not require depth confirmation (operators can still waive it per job)")
args = ap.parse_args()
analyze_fn, info = model_analyzer(args.checkpoint, args.device)
console = Console(args.jobs_dir, analyze_fn, info,
{"mm_per_px": args.mm_per_px, "require_depth": not args.no_require_depth,
"window_gate": not args.no_window_gate})
server = ThreadingHTTPServer((args.host, args.port), make_handler(console))
for warning in info["warnings"]:
print("warning:", warning)
print(f"console on http://{args.host}:{args.port} (model {info['checkpoint']}, jobs in {args.jobs_dir})", flush=True)
try:
server.serve_forever()
except KeyboardInterrupt:
pass
if __name__ == "__main__":
main()
|