File size: 5,175 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
"""Before/after gallery on held-out RAQEEB test images, with a per-image check against the ground truth.



  python make_gallery.py --weights ... --thresholds ... --out <dir>



Images are chosen by class content (most distinct annotated classes per source), never by model performance.

Each image is scored only on the classes its source fully annotates (dataset/label_sets.json): a predicted box

counts as correct when it overlaps a ground-truth box of the same class with IoU >= 0.5 (one-to-one).

"""
import argparse
import glob
import json
from pathlib import Path
from types import SimpleNamespace

import numpy as np
from PIL import Image

from annotate_raqeeb import NAMES, before_after, boxes_from_detections, draw_boxes

RAQ = Path(r'D:\CCTV_AI_FLOORPLAN_data\RAQEEB CCTV')
GALLERY = [  # (id prefix, file name, source description, licence)
    ('RES-01__high_quality_architectural_416', '01_cubicasa_architectural_plan',
     'CubiCasa5K (high_quality_architectural/416)', 'CC BY-NC 4.0'),
    ('RES-02__train-01_1031-0029', '02_floorplancad_cad_sheet', 'FloorPlanCAD (CAD drawing crop)', 'CC BY-NC 4.0'),
    ('RES-02__test-00_0067-0021', '03_floorplancad_escalator', 'FloorPlanCAD (CAD drawing crop)', 'CC BY-NC 4.0'),
    ('RF-09__test_ROW_FP_SPA_843441212', '04_lifts_stairs_spaces', 'Roboflow Universe kt/floor-spac', 'CC BY 4.0'),
    ('RF-12__test_evctmap-413', '05_evacuation_plan_exits', 'Roboflow Universe tw/evacuation-plan', 'CC BY 4.0'),
    ('RF-02__train_Working-drawings_stamped-pdf_14', '06_working_drawing_sheet', 'Roboflow Universe Doxle/floor-plan',
     'CC BY 4.0'),
    ('RF-01__test_1235_png_jpg', '07_residential_plan', 'Roboflow Universe Floorplanes/training-ai', 'CC BY 4.0'),
    ('RF-14__test_FloorPlan-13-', '08_simple_floor_plan', 'Roboflow Universe simple-floor-plan/floor-plan',
     'CC BY 4.0'),
]


def gt_boxes(label_path, w, h):
    out = []
    for line in Path(label_path).read_text().splitlines():
        v = line.split()
        if len(v) < 7:
            continue
        xy = np.array(v[1:], float).reshape(-1, 2) * [w, h]
        out.append((int(v[0]), [*xy.min(0), *xy.max(0)]))
    return out


def iou(a, b):
    ix = max(0, min(a[2], b[2]) - max(a[0], b[0]))
    iy = max(0, min(a[3], b[3]) - max(a[1], b[1]))
    i = ix * iy
    return i / max((a[2] - a[0]) * (a[3] - a[1]) + (b[2] - b[0]) * (b[3] - b[1]) - i, 1e-9)


def main():
    ap = argparse.ArgumentParser()
    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('--max-side', type=int, default=1400)
    ap.add_argument('--device', default='0')
    a = ap.parse_args()
    from ultralytics import YOLO
    from infer_raqeeb import infer_page
    a.out.mkdir(parents=True, exist_ok=True)
    sets = json.loads((RAQ / 'dataset' / 'label_sets.json').read_text())
    model = YOLO(str(a.weights))
    config = json.loads(a.thresholds.read_text())
    args = SimpleNamespace(tile=0, overlap=256, imgsz=1024, device=a.device)
    report = []
    for prefix, fname, source, licence in GALLERY:
        img = sorted(glob.glob(str(RAQ / 'dataset' / 'images' / 'test' / f'{prefix}*')))[0]
        iid = Path(img).stem
        im = Image.open(img).convert('RGB')
        dets, _ = infer_page(model, im, config, args)
        boxes = boxes_from_detections(dets)
        full = sets['image_full_overrides'].get(iid, sets['sources'][iid.split('__')[0]]['full'])
        gts = gt_boxes(RAQ / 'dataset' / 'labels' / 'test' / f'{iid}.txt', *im.size)
        per = {}
        for c in sorted(full):
            g = [b for k, b in gts if k == c]
            p = sorted([b for b in boxes if b['class_id'] == c], key=lambda b: -b['confidence'])
            if not g and not p:
                continue
            used, tp = set(), 0
            for b in p:
                best = max(((iou(b['box_xyxy'], x), j) for j, x in enumerate(g) if j not in used), default=(0, None))
                if best[0] >= 0.5:
                    used.add(best[1])
                    tp += 1
            per[NAMES[c]] = dict(ground_truth=len(g), predicted=len(p), correct_iou50=tp)
        s = min(1.0, a.max_side / max(im.size))
        small = im.resize((round(im.width * s), round(im.height * s)), Image.LANCZOS) if s < 1 else im
        after = draw_boxes(small, [dict(b, box_xyxy=[v * s for v in b['box_xyxy']]) for b in boxes])
        before_after(small, after).save(a.out / f'{fname}.jpg', quality=88)
        report.append(dict(file=f'{fname}.jpg', image_id=iid, source=source, licence=licence, split='test',
                           scored_classes=[NAMES[c] for c in sorted(full)], per_class=per,
                           all_predictions={n: sum(1 for b in boxes if b['label'] == n)
                                            for n in sorted({b['label'] for b in boxes})}))
        print(fname, json.dumps(per), flush=True)
    (a.out / 'gallery.json').write_text(json.dumps(report, indent=1))


if __name__ == '__main__':
    main()