File size: 6,746 Bytes
d70361b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Prepare hand-labeling assets for the DDA pair Grid_54.tif vs H43X2E1.tif.

Creates a paint/QGIS-friendly pack under ``docs/delhi_eval/dda_labeling/``:
  - before.png / after.png   (downscaled RGB, default max_side=4096)
  - seed_mask.png            (current AdaptFormer+recovery detection as a draft)
  - gt_mask_blank.png        (empty mask — paint change as white on black)
  - LABELING.md              (how to finish the label and ingest it)
  - meta.json

Also registers the pair in ``docs/delhi_eval/manifest.json`` as ``dda_grid54_h43x2e1``.

Usage (from repo root):
  python scripts/prepare_dda_gt_labeling.py
  python scripts/prepare_dda_gt_labeling.py --max-side 3072 --skip-seed
"""
from __future__ import annotations

import argparse
import json
import sys
import time
from pathlib import Path

import numpy as np
from PIL import Image

ROOT = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(ROOT))

BEFORE = ROOT / "data/library_sources/central_delhi/Images/Grid_54.tif"
AFTER = ROOT / "data/library_sources/central_delhi/Images/H43X2E1.tif"
OUT_DIR = ROOT / "docs/delhi_eval/dda_labeling/dda_grid54_h43x2e1"
PAIR_ID = "dda_grid54_h43x2e1"
MANIFEST = ROOT / "docs/delhi_eval/manifest.json"
LABEL_PATH = ROOT / "docs/delhi_eval/labels" / f"{PAIR_ID}.png"


def _load_rgb(path: Path, max_side: int) -> np.ndarray:
    from app.dda.geotiff_io import load_rgb_pil
    return np.array(load_rgb_pil(path, max_side=max_side).convert("RGB"))


def _register_manifest(preview_w: int, preview_h: int) -> None:
    data = json.loads(MANIFEST.read_text(encoding="utf-8"))
    pairs = data.get("pairs") or []
    entry = {
        "pair_id": PAIR_ID,
        "before_path": str(BEFORE.relative_to(ROOT)).replace("\\", "/"),
        "after_path": str(AFTER.relative_to(ROOT)).replace("\\", "/"),
        "date_before": None,
        "date_after": None,
        "gsd": 0.03,
        "zone": "Central Delhi / Grid 54",
        "change_types": ["building", "road", "vegetation", "mixed_gsd"],
        "gt_mask": None,
        "notes": (
            "DDA production pair (Grid_54 vs H43X2E1). Hand-label in "
            f"docs/delhi_eval/dda_labeling/{PAIR_ID}/ then run "
            "scripts/ingest_dda_gt_label.py"
        ),
        "label_preview_size": [preview_w, preview_h],
    }
    # replace if exists
    pairs = [p for p in pairs if p.get("pair_id") != PAIR_ID]
    pairs.append(entry)
    data["pairs"] = pairs
    MANIFEST.write_text(json.dumps(data, indent=2), encoding="utf-8")
    print(f"Registered {PAIR_ID} in manifest ({len(pairs)} pairs total)")


def _write_guide(out: Path) -> None:
    text = f"""# Labeling pack: Grid_54 vs H43X2E1

## Goal
Hand-draw **real permanent ground change** (new buildings, demolition, roads).
Do **not** mark cars, shadows, seasonal tree canopy, or illumination shifts.

## Files
| File | Use |
|---|---|
| `before.png` | earlier date |
| `after.png` | later date |
| `seed_mask.png` | draft from current detector (white=change) — **edit this** |
| `gt_mask_blank.png` | empty alternative if you prefer starting from scratch |

## Paint / Photoshop
1. Open `after.png` and `seed_mask.png`.
2. Edit the mask so **white (255) = true change**, **black (0) = no change**.
3. Save as `gt_mask.png` in this folder (single-channel or RGB white/black).

## QGIS
1. Load `before.png` / `after.png` as rasters (same extent).
2. Digitize change polygons → rasterize to same size as the PNGs → export
   binary GeoTIFF/PNG as `gt_mask.png`.

## Ingest into the eval set
```bash
python scripts/ingest_dda_gt_label.py
```
This copies `gt_mask.png` → `docs/delhi_eval/labels/{PAIR_ID}.png` and updates
the manifest. Calibration / fine-tune scripts will then use it.

## Notes
- Preview size is downscaled from native ~25k×28k for tractable labeling.
- After ingest, re-run calibration:
  `python scripts/v3_error_analysis_and_thr_sweep.py`
  `python scripts/grid_search_calibration.py --methods "AI-Based Deep Learning" --sensitivities 0.4,0.5,0.6 --pair-ids {PAIR_ID}`
"""
    (out / "LABELING.md").write_text(text, encoding="utf-8")


def main() -> int:
    ap = argparse.ArgumentParser()
    ap.add_argument("--max-side", type=int, default=4096)
    ap.add_argument("--skip-seed", action="store_true",
                    help="Skip AdaptFormer seed mask (faster pack-only)")
    args = ap.parse_args()

    if not BEFORE.is_file() or not AFTER.is_file():
        print("Missing Grid_54.tif or H43X2E1.tif under data/library_sources/...")
        return 1

    OUT_DIR.mkdir(parents=True, exist_ok=True)
    print(f"Loading pair at max_side={args.max_side} ...")
    t0 = time.time()
    before = _load_rgb(BEFORE, args.max_side)
    after = _load_rgb(AFTER, args.max_side)
    print(f"  loaded {before.shape} in {time.time()-t0:.1f}s")

    Image.fromarray(before).save(OUT_DIR / "before.png")
    Image.fromarray(after).save(OUT_DIR / "after.png")
    blank = np.zeros(before.shape[:2], dtype=np.uint8)
    Image.fromarray(blank).save(OUT_DIR / "gt_mask_blank.png")

    seed_px = 0
    if not args.skip_seed:
        print("Running detection for seed mask (this can take several minutes)...")
        from dotenv import load_dotenv
        load_dotenv(ROOT / ".env", override=True)
        from app.detection_engine import run_detection
        mask, _vis, stats, regions = run_detection(
            Image.fromarray(before),
            Image.fromarray(after),
            method="AI-Based Deep Learning",
            enable_registration=True,
            enable_normalization=True,
            detection_sensitivity=0.5,
            max_size=args.max_side,
            before_path=str(BEFORE),
            after_path=str(AFTER),
        )
        seed = np.array(mask)
        if seed.ndim == 3:
            seed = seed[:, :, 0]
        Image.fromarray(seed.astype(np.uint8)).save(OUT_DIR / "seed_mask.png")
        seed_px = int((seed > 127).sum())
        print(f"  seed change%={stats.get('change_percentage')} regions={len(regions)} px={seed_px}")
    else:
        Image.fromarray(blank).save(OUT_DIR / "seed_mask.png")

    meta = {
        "pair_id": PAIR_ID,
        "before": str(BEFORE),
        "after": str(AFTER),
        "preview_shape": list(before.shape[:2]),
        "max_side": args.max_side,
        "seed_changed_px": seed_px,
        "created_unix": time.time(),
    }
    (OUT_DIR / "meta.json").write_text(json.dumps(meta, indent=2), encoding="utf-8")
    _write_guide(OUT_DIR)
    _register_manifest(before.shape[1], before.shape[0])
    print(f"Labeling pack ready: {OUT_DIR}")
    print("Next: edit seed_mask.png → save as gt_mask.png → run scripts/ingest_dda_gt_label.py")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())