Image Segmentation
ultralytics
yolo
yolo11
instance-segmentation
object-detection
floor-plan
architecture
cad
autocad
building-elements
cctv
Eval Results (legacy)
Instructions to use swedishtech/RAQEEB-CCTV with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- ultralytics
How to use swedishtech/RAQEEB-CCTV with ultralytics:
from huggingface_hub import hf_hub_download from ultralytics import YOLO # pick the weights file from this repo's "Files and versions" tab weights = hf_hub_download("swedishtech/RAQEEB-CCTV", "<weights>.pt") model = YOLO(weights) source = 'http://images.cocodataset.org/val2017/000000039769.jpg' model.predict(source=source, save=True) - Notebooks
- Google Colab
- Kaggle
File size: 19,873 Bytes
d0f2b17 | 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 | """RAQEEB on AutoCAD drawings (DWG or DXF): labelled boxes on the rendered drawing and inside the drawing itself.
python cad_raqeeb.py drawing.dwg --weights raqeeb-cctv-v1.0.0.pt --thresholds thresholds.json --out out_dir ^
--region ground=194976,1278147,273726,1365497 [--region ...] [--px-per-unit 0.1] [--plot] [--tarch-check]
Steps
1. DWG -> DXF with AutoCAD's headless accoreconsole (a DXF input is used as is).
2. Each region (drawing units) is rendered black on white at --px-per-unit pixels per drawing unit; 0.1 means
100 px per metre for a millimetre drawing, the scale of the CAD (FloorPlanCAD) training images.
3. RAQEEB runs in overlapping --tile px tiles with the validation-selected per-class thresholds (the same
infer_raqeeb.infer_page used for PNG/JPEG/PDF) and every detection is drawn as a box with its class name.
4. Each box is mapped back to drawing coordinates and added to a copy of the drawing, one layer per class
(RAQEEB-WALL, RAQEEB-DOOR, ...): a closed rectangle plus a 'door 0.91' text -> <name>_raqeeb.dxf and, through
AutoCAD, <name>_raqeeb.dwg.
5. --plot: AutoCAD plots each region of the annotated DWG to PDF (DWG To PDF.pc3), i.e. the boxes and labels as
AutoCAD itself draws them.
6. --tarch-check: for drawings made with TArch (layers WINDOW, COLUMN, WALL, STAIR, SPACE) the detections are scored
against the drawing's own objects: door/window/column blocks, wall lines, stair lines and room-name texts.
"""
import argparse
import glob
import json
import re
import subprocess
import time
from pathlib import Path
from types import SimpleNamespace
import numpy as np
from PIL import Image
from annotate_raqeeb import COLORS, NAMES, before_after, boxes_from_detections, draw_boxes
Image.MAX_IMAGE_PIXELS = 400_000_000
ACI = [1, 3, 5, 30, 6, 4, 200, 40, 221, 94, 150] # fallback AutoCAD colour index per class
# ----------------------------------------------------------------------------------------------- AutoCAD
def find_accoreconsole():
hits = sorted(glob.glob(r'C:\Program Files\Autodesk\AutoCAD 20*\accoreconsole.exe'))
if not hits:
raise FileNotFoundError('AutoCAD accoreconsole.exe not found; pass --accore or give a DXF')
return hits[-1]
def acad(accore, infile, lines, workdir, tag):
"""Run an AutoCAD script headless on `infile`; returns the console log text."""
scr = workdir / f'_{tag}.scr'
scr.write_text('\n'.join(lines) + '\n', encoding='ascii')
r = subprocess.run([accore, '/i', str(infile), '/s', str(scr), '/l', 'en-US'], cwd=str(workdir),
capture_output=True, timeout=3600)
log = (r.stdout + r.stderr).replace(b'\x00', b'').decode('utf-8', 'replace')
(workdir / f'_{tag}.log').write_text(log, encoding='utf-8')
return log
def dwg_to_dxf(accore, dwg, dxf, workdir):
dxf.unlink(missing_ok=True)
acad(accore, dwg, ['_.FILEDIA', '0', '_.CMDDIA', '0', '_.SAVEAS', 'DXF', 'V', '2018', '16',
f'"{dxf.as_posix()}"', '_.QUIT', 'Y'], workdir, 'dwg_to_dxf')
if not dxf.exists():
raise RuntimeError(f'AutoCAD did not write {dxf}; see {workdir / "_dwg_to_dxf.log"}')
def dxf_to_dwg(accore, dxf, dwg, workdir):
dwg.unlink(missing_ok=True)
acad(accore, dxf, ['_.FILEDIA', '0', '_.CMDDIA', '0', '_.SAVEAS', '2018', f'"{dwg.as_posix()}"',
'_.QUIT', 'Y'], workdir, 'dxf_to_dwg')
if not dwg.exists():
raise RuntimeError(f'AutoCAD did not write {dwg}; see {workdir / "_dxf_to_dwg.log"}')
def plot_pdf(accore, dwg, region, pdf, workdir, tag):
"""Plot a model-space window of `dwg` to PDF with AutoCAD (object colours, lineweights)."""
x0, y0, x1, y1 = region
portrait = (y1 - y0) > (x1 - x0)
pdf.unlink(missing_ok=True)
acad(accore, dwg, [
'_.FILEDIA', '0', '_.CMDDIA', '0', '_.BACKGROUNDPLOT', '0',
'-PLOT', 'Y', 'Model', 'DWG To PDF.pc3', 'ISO full bleed A1 (841.00 x 594.00 MM)', 'M',
'P' if portrait else 'L', 'N', 'W', f'{x0},{y0}', f'{x1},{y1}', 'F', 'C', 'Y', '.', 'Y', 'A',
f'"{pdf.as_posix()}"', 'N', 'Y', '_.QUIT', 'Y'], workdir, tag)
return pdf.exists()
def pdf_to_png(pdf, png, max_side=5000):
import pypdfium2 as pdfium
doc = pdfium.PdfDocument(str(pdf))
page = doc[0]
scale = max_side / max(page.get_width(), page.get_height())
page.render(scale=scale).to_pil().convert('RGB').save(png)
doc.close()
# ----------------------------------------------------------------------------------------------- rendering
def load_for_render(dxf, font):
import ezdxf
from ezdxf import bbox
doc = ezdxf.readfile(dxf)
for s in doc.styles: # SHX/big fonts are not available outside AutoCAD; use one TrueType CJK-capable font
s.dxf.font = font
s.dxf.bigfont = ''
cache = bbox.Cache()
ents = []
for e in doc.modelspace():
b = bbox.extents([e], fast=True, cache=cache)
if b.has_data:
ents.append((e, (b.extmin.x, b.extmin.y, b.extmax.x, b.extmax.y)))
return doc, ents
def render(doc, ents, region, ppu, out_png, lineweight):
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from ezdxf.addons.drawing import Frontend, RenderContext
from ezdxf.addons.drawing.config import BackgroundPolicy, ColorPolicy, Configuration
from ezdxf.addons.drawing.matplotlib import MatplotlibBackend
x0, y0, x1, y1 = region
W, H = round((x1 - x0) * ppu), round((y1 - y0) * ppu)
sel = [e for e, b in ents if b[2] >= x0 and b[0] <= x1 and b[3] >= y0 and b[1] <= y1]
fig = plt.figure(figsize=(W / 100, H / 100), dpi=100)
ax = fig.add_axes([0, 0, 1, 1])
ax.set_axis_off()
cfg = Configuration(color_policy=ColorPolicy.BLACK, background_policy=BackgroundPolicy.WHITE,
lineweight_scaling=0, min_lineweight=lineweight)
Frontend(RenderContext(doc), MatplotlibBackend(ax), config=cfg).draw_entities(sel)
ax.set_xlim(x0, x1)
ax.set_ylim(y0, y1)
fig.savefig(out_png, dpi=100, facecolor='white')
plt.close(fig)
im = Image.open(out_png).convert('RGB')
if im.size != (W, H):
im = im.resize((W, H), Image.LANCZOS)
im.save(out_png)
return im, len(sel)
# ----------------------------------------------------------------------------------------------- geometry
def px_to_world(box, region, ppu):
x0, y0, x1, y1 = box
rx0, _, _, ry1 = region
return [rx0 + x0 / ppu, ry1 - y1 / ppu, rx0 + x1 / ppu, ry1 - y0 / ppu]
def world_to_px(pt, region, ppu):
return ((pt[0] - region[0]) * ppu, (region[3] - pt[1]) * ppu)
def write_annotated(dxf_in, dxf_out, dets, text_h):
import ezdxf
doc = ezdxf.readfile(dxf_in)
msp = doc.modelspace()
if 'RAQEEB' not in doc.styles:
doc.styles.new('RAQEEB', dxfattribs={'font': 'arial.ttf'})
for c, n in enumerate(NAMES):
name = f'RAQEEB-{n.upper()}'
if name not in doc.layers:
lay = doc.layers.add(name, color=ACI[c])
lay.rgb = COLORS[c]
lay.dxf.lineweight = 35
for d in dets:
layer = f"RAQEEB-{d['label'].upper()}"
X0, Y0, X1, Y1 = d['box_world']
msp.add_lwpolyline([(X0, Y0), (X1, Y0), (X1, Y1), (X0, Y1)], close=True, dxfattribs={'layer': layer})
t = msp.add_text(f"{d['label']} {d['confidence']:.2f}", height=text_h,
dxfattribs={'layer': layer, 'style': 'RAQEEB'})
t.set_placement((X0, Y1 + 0.25 * text_h))
doc.saveas(dxf_out)
# ----------------------------------------------------------------------------------------------- TArch check
DOOR_BLOCK = re.compile(r'DOOR|DorLib|LDOR', re.I)
WIN_BLOCK = re.compile(r'WIN', re.I)
def tarch_truth(ents, region):
x0, y0, x1, y1 = region
inside = lambda b: x0 <= (b[0] + b[2]) / 2 <= x1 and y0 <= (b[1] + b[3]) / 2 <= y1
gt = dict(door=[], window=[], column=[], wall_lines=[], stair_lines=[], room_points=[])
for e, b in ents:
if not inside(b):
continue
t, layer = e.dxftype(), e.dxf.layer.upper()
if t == 'INSERT':
name = e.dxf.name
if WIN_BLOCK.search(name):
gt['window'].append(b)
elif DOOR_BLOCK.search(name):
gt['door'].append(b)
elif layer == 'COLUMN':
gt['column'].append(b)
elif t == 'LINE' and layer == 'WALL':
gt['wall_lines'].append((e.dxf.start.x, e.dxf.start.y, e.dxf.end.x, e.dxf.end.y))
elif t in ('LINE', 'LWPOLYLINE') and layer in ('STAIR', '梯道'):
pts = [(e.dxf.start.x, e.dxf.start.y), (e.dxf.end.x, e.dxf.end.y)] if t == 'LINE' else \
[(p[0], p[1]) for p in e.get_points('xy')]
gt['stair_lines'].extend((*pts[i], *pts[i + 1]) for i in range(len(pts) - 1))
elif t == 'TEXT' and layer == 'SPACE' and e.dxf.text.strip():
p = e.dxf.insert
gt['room_points'].append((p.x, p.y))
return gt
def _iou_ios(p, g):
ix = max(0.0, min(p[2], g[2]) - max(p[0], g[0]))
iy = max(0.0, min(p[3], g[3]) - max(p[1], g[1]))
inter = ix * iy
ap, ag = (p[2] - p[0]) * (p[3] - p[1]), (g[2] - g[0]) * (g[3] - g[1])
return inter / max(ap + ag - inter, 1e-9), inter / max(min(ap, ag), 1e-9)
def match(preds, gts, which, thr=0.5):
"""Greedy one-to-one matching by confidence; which=0 -> IoU, 1 -> intersection over the smaller box."""
used, tp = set(), 0
for p in sorted(preds, key=lambda d: -d['confidence']):
best, bv = None, thr
for j, g in enumerate(gts):
if j in used:
continue
v = _iou_ios(p['box_world'], g)[which]
if v >= bv:
best, bv = j, v
if best is not None:
used.add(best)
tp += 1
return tp
def prf(tp, n_pred, n_gt):
p = tp / n_pred if n_pred else None
r = tp / n_gt if n_gt else None
f = 2 * p * r / (p + r) if p and r else (0.0 if p is not None and r is not None else None)
return dict(true_positives=tp, predictions=n_pred, truth=n_gt, precision=None if p is None else round(p, 3),
recall=None if r is None else round(r, 3), f1=None if f is None else round(f, 3))
def tarch_check(gt, dets, raw_dets, region, ppu, size):
import cv2
from shapely.geometry import Point
from shapely.ops import unary_union
W, H = size
out = {}
by = lambda n: [d for d in dets if d['label'] == n]
for n in ('door', 'window', 'column'):
out[n] = dict(IoU_0_5=prf(match(by(n), gt[n], 0), len(by(n)), len(gt[n])),
overlap_0_5=prf(match(by(n), gt[n], 1), len(by(n)), len(gt[n])))
# walls: pixel agreement between predicted wall masks and the drawing's WALL lines
lines = np.zeros((H, W), np.uint8)
for x0, y0, x1, y1 in gt['wall_lines']:
a, b = world_to_px((x0, y0), region, ppu), world_to_px((x1, y1), region, ppu)
cv2.line(lines, tuple(int(round(v)) for v in a), tuple(int(round(v)) for v in b), 1, 1)
pred = np.zeros((H, W), np.uint8)
for d in raw_dets:
if d['class_id'] != 0:
continue
geoms = getattr(d['geometry'], 'geoms', [d['geometry']])
for g in geoms:
if g.geom_type == 'Polygon' and not g.is_empty:
cv2.fillPoly(pred, [np.round(np.asarray(g.exterior.coords)).astype(np.int32)], 1)
if lines.any() and pred.any():
tol_r, tol_p = max(1, round(50 * ppu)), max(1, round(250 * ppu))
near_pred = cv2.dilate(pred, np.ones((2 * tol_r + 1, 2 * tol_r + 1), np.uint8))
dist = cv2.distanceTransform((1 - lines).astype(np.uint8), cv2.DIST_L2, 3)
out['wall'] = dict(line_pixels_covered=round(float(near_pred[lines > 0].mean()), 3),
predicted_pixels_on_walls=round(float((dist[pred > 0] <= tol_p).mean()), 3),
note='recall-like: share of WALL-layer line pixels within 50 units of a predicted '
'wall mask; precision-like: share of predicted wall pixels within 250 units of a '
'WALL-layer line (drawing units, mm here)')
# stairs: connected groups of STAIR-layer lines -> boxes
st = np.zeros((H, W), np.uint8)
for x0, y0, x1, y1 in gt['stair_lines']:
a, b = world_to_px((x0, y0), region, ppu), world_to_px((x1, y1), region, ppu)
cv2.line(st, tuple(int(round(v)) for v in a), tuple(int(round(v)) for v in b), 1, 1)
k = max(3, round(400 * ppu)) | 1
st = cv2.morphologyEx(st, cv2.MORPH_CLOSE, np.ones((k, k), np.uint8))
n, _, stats, _ = cv2.connectedComponentsWithStats(st)
stair_gt = []
for i in range(1, n):
x, y, w, h = stats[i, :4]
if min(w, h) >= 800 * ppu:
stair_gt.append(px_to_world((x, y, x + w, y + h), region, ppu))
out['stair'] = dict(overlap_0_5=prf(match(by('stair'), stair_gt, 1), len(by('stair')), len(stair_gt)))
# rooms: are the room-name texts inside a predicted space_region?
spaces = [d['geometry'] for d in raw_dets if d['class_id'] == 10]
if gt['room_points']:
u = unary_union(spaces) if spaces else None
hit = sum(1 for p in gt['room_points'] if u is not None and u.contains(Point(*world_to_px(p, region, ppu))))
out['space_region'] = dict(room_labels=len(gt['room_points']), inside_a_predicted_space=hit,
share=round(hit / len(gt['room_points']), 3), predicted_spaces=len(spaces))
out['truth_definition'] = ('door: TArch door blocks (*DOOR*, DorLib*, LDOR*), window: *WIN* blocks, column: '
'blocks on layer COLUMN, stair: connected STAIR/梯道 line groups, rooms: SPACE texts. '
'IoU_0_5 = box IoU >= 0.5; overlap_0_5 = intersection over the smaller box >= 0.5 '
'(tolerates door swing arcs being in or out of a box).')
return out
# ----------------------------------------------------------------------------------------------- main
def main():
ap = argparse.ArgumentParser()
ap.add_argument('drawing', type=Path)
ap.add_argument('--weights', type=Path, required=True)
ap.add_argument('--thresholds', type=Path, required=True)
ap.add_argument('--out', type=Path, required=True)
ap.add_argument('--region', action='append', default=[], help='name=x0,y0,x1,y1 in drawing units')
ap.add_argument('--px-per-unit', type=float, default=0.1)
ap.add_argument('--tile', type=int, default=1024)
ap.add_argument('--overlap', type=int, default=256)
ap.add_argument('--imgsz', type=int, default=1024)
ap.add_argument('--device', default='0')
ap.add_argument('--lineweight', type=float, default=4, help='rendered line width in 1/300 inch at 100 dpi')
ap.add_argument('--font', default='msyh.ttc')
ap.add_argument('--text-height', type=float, help='label height in drawing units (default: 0.25%% of region)')
ap.add_argument('--plot', action='store_true', help='plot the annotated DWG regions to PDF with AutoCAD')
ap.add_argument('--tarch-check', action='store_true')
ap.add_argument('--no-dwg', action='store_true', help='skip writing the annotated DWG')
ap.add_argument('--accore', help='path to accoreconsole.exe')
a = ap.parse_args()
a.out.mkdir(parents=True, exist_ok=True)
out = a.out.resolve()
accore = None
if a.drawing.suffix.lower() == '.dwg' or a.plot or not a.no_dwg:
accore = a.accore or find_accoreconsole()
stem = a.drawing.stem
if a.drawing.suffix.lower() == '.dwg':
dxf = out / f'{stem}.dxf'
if not dxf.exists():
print('DWG -> DXF with', accore, flush=True)
dwg_to_dxf(accore, a.drawing.resolve(), dxf, out)
else:
dxf = a.drawing.resolve()
t = time.time()
doc, ents = load_for_render(dxf, a.font)
print(f'loaded {len(ents)} entities in {time.time() - t:.0f}s', flush=True)
regions = []
for spec in a.region or []:
name, coords = spec.split('=')
regions.append((name, [float(v) for v in coords.split(',')]))
if not regions:
xs = [b for _, b in ents]
regions = [('extents', [min(b[0] for b in xs), min(b[1] for b in xs), max(b[2] for b in xs),
max(b[3] for b in xs)])]
from ultralytics import YOLO
from infer_raqeeb import infer_page
model = YOLO(str(a.weights))
config = json.loads(a.thresholds.read_text())
args = SimpleNamespace(tile=a.tile, overlap=a.overlap, imgsz=a.imgsz, device=a.device)
ppu = a.px_per_unit
all_dets, summary = [], dict(drawing=str(a.drawing), dxf=str(dxf), px_per_unit=ppu, tile=a.tile,
overlap=a.overlap, thresholds=str(a.thresholds), weights=str(a.weights), regions={})
for name, region in regions:
t = time.time()
im, n_ent = render(doc, ents, region, ppu, out / f'{name}_input.png', a.lineweight)
t_render = time.time() - t
raw, _ = infer_page(model, im, config, args)
t_det = time.time() - t - t_render
boxes = boxes_from_detections(raw)
for b in boxes:
b['box_world'] = [round(v, 1) for v in px_to_world(b['box_xyxy'], region, ppu)]
all_dets.extend(boxes)
draw_boxes(im, boxes, line=2, font_px=15).save(out / f'{name}_annotated_full.jpg', quality=90)
s = min(1.0, 2400 / max(im.size))
small = im.resize((round(im.width * s), round(im.height * s)), Image.LANCZOS)
after = draw_boxes(small, [dict(b, box_xyxy=[v * s for v in b['box_xyxy']]) for b in boxes], font_px=12)
after.save(out / f'{name}_annotated.jpg', quality=92)
before_after(small, after, title_left='AutoCAD drawing (rendered)').save(out / f'{name}_before_after.jpg',
quality=90)
counts = {}
for b in boxes:
counts[b['label']] = counts.get(b['label'], 0) + 1
rec = dict(region=region, image_size=list(im.size), entities_rendered=n_ent, counts=counts,
seconds_render=round(t_render, 1), seconds_detect=round(t_det, 1))
if a.tarch_check:
rec['tarch_check'] = tarch_check(tarch_truth(ents, region), boxes, raw, region, ppu, im.size)
summary['regions'][name] = rec
(out / f'{name}_detections.json').write_text(json.dumps(dict(rec, boxes=boxes), indent=1, ensure_ascii=False),
encoding='utf-8')
print(f'{name}: {len(boxes)} boxes {counts} render {t_render:.0f}s detect {t_det:.0f}s', flush=True)
if a.tarch_check:
print(json.dumps(rec['tarch_check'], ensure_ascii=False), flush=True)
text_h = a.text_height or 0.0025 * max(max(r[2] - r[0], r[3] - r[1]) for _, r in regions)
ann_dxf = out / f'{stem}_raqeeb.dxf'
write_annotated(dxf, ann_dxf, all_dets, text_h)
summary['annotated_dxf'] = str(ann_dxf)
if not a.no_dwg:
ann_dwg = out / f'{stem}_raqeeb.dwg'
dxf_to_dwg(accore, ann_dxf, ann_dwg, out)
summary['annotated_dwg'] = str(ann_dwg)
if a.plot:
for name, region in regions:
pdf = out / f'{name}_autocad_plot.pdf'
ok = plot_pdf(accore, ann_dwg, region, pdf, out, f'plot_{name}')
if ok:
pdf_to_png(pdf, out / f'{name}_autocad_plot.png')
summary['regions'][name]['autocad_plot'] = str(pdf) if ok else 'failed (see log)'
print(f'{name}: AutoCAD plot', 'ok' if ok else 'FAILED', flush=True)
(out / 'summary.json').write_text(json.dumps(summary, indent=2, ensure_ascii=False), encoding='utf-8')
print('done ->', out)
if __name__ == '__main__':
main()
|