Spaces:
Running
Running
File size: 6,041 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 | """Ablate before6/after6 using the live GeoTIFF job path (resize + skip-reg NCC)."""
from __future__ import annotations
import json
import os
import sys
from pathlib import Path
import cv2
import numpy as np
from PIL import Image
from dotenv import load_dotenv
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT))
os.chdir(ROOT)
load_dotenv(ROOT / ".env", override=True)
from app.detection_engine import ( # noqa: E402
_alignment_ncc,
ai_deep_learning_method,
analyze_change_regions,
get_detection_max_size,
normalize_radiometry,
preprocess_image,
recover_chromatic_roof_construction,
recover_dark_roof_construction,
strip_parking_cluster_from_mask,
strip_shadow_fragments_from_mask,
strip_shadow_only_from_mask,
strip_transient_from_mask,
strip_weak_seasonal_veg_from_mask,
)
def count(m):
return int(np.sum(m > 127))
def main():
before_p = ROOT / "data/library_sources/central_delhi/Images/before6.tif"
after_p = ROOT / "data/library_sources/central_delhi/Images/after6.tif"
before_pil = Image.open(before_p).convert("RGB")
after_pil = Image.open(after_p).convert("RGB")
# Match job_runner: resize after onto before
if after_pil.size != before_pil.size:
after_pil = after_pil.resize(before_pil.size, Image.Resampling.LANCZOS)
ms = get_detection_max_size()
b = preprocess_image(before_pil, max_size=ms)
a = preprocess_image(after_pil, max_size=ms)
ncc = float(_alignment_ncc(b, a))
registration_ok = bool(ncc >= 0.55)
print("shape", b.shape, "ncc", round(ncc, 4), "registration_ok", registration_ok)
print("weights", os.environ.get("ADAPTFORMER_WEIGHTS"), "thr", os.environ.get("ADAPTFORMER_THRESHOLD"))
b_chr, a_chr = b.copy(), a.copy()
b, a = normalize_radiometry(b, a)
change_mask, debug = ai_deep_learning_method(
b, a, sensitivity=0.5, registration_ok=registration_ok
)
print("dl_debug", {k: debug.get(k) for k in (
"threshold_score", "model_changed_px", "combined_changed_px", "method", "model"
) if isinstance(debug, dict)})
steps = [("raw_dl", change_mask.copy())]
change_mask = strip_transient_from_mask(change_mask, b, a)
steps.append(("strip_transient", change_mask.copy()))
change_mask = strip_shadow_only_from_mask(change_mask, b, a)
steps.append(("strip_shadow_only", change_mask.copy()))
change_mask = strip_shadow_fragments_from_mask(
change_mask, b, a, registration_ok=registration_ok
)
steps.append(("strip_shadow_fragments_soft", change_mask.copy()))
change_mask = strip_parking_cluster_from_mask(change_mask, b, a)
steps.append(("strip_parking", change_mask.copy()))
change_mask = strip_weak_seasonal_veg_from_mask(change_mask, b, a)
steps.append(("strip_weak_veg", change_mask.copy()))
change_mask = recover_chromatic_roof_construction(change_mask, b_chr, a_chr)
steps.append(("recover_chromatic", change_mask.copy()))
change_mask = recover_dark_roof_construction(change_mask, b_chr, a_chr)
steps.append(("recover_dark_roof", change_mask.copy()))
change_mask = strip_shadow_only_from_mask(change_mask, b, a)
steps.append(("re_strip_shadow_only", change_mask.copy()))
change_mask = strip_shadow_fragments_from_mask(
change_mask, b, a, registration_ok=registration_ok
)
steps.append(("re_strip_shadow_fragments_soft", change_mask.copy()))
out = ROOT / "data/delhi_cd/friday_drone_report_fix/run73_ablation"
out.mkdir(parents=True, exist_ok=True)
prev = None
rows = []
for name, m in steps:
px = count(m)
delta = None if prev is None else px - prev
rows.append({"step": name, "px": px, "delta": delta, "pct": round(100 * px / m.size, 3)})
print(f"{name:32s} px={px:7d} delta={str(delta):>8s} pct={100*px/m.size:.3f}%")
cv2.imwrite(str(out / f"{name}.png"), m)
prev = px
regs = analyze_change_regions(
change_mask, a_chr, min_area=150, use_ensemble=False, before_img=b_chr,
registration_ok=registration_ok,
)
print("final_regions", len(regs), "top_areas", [r["area"] for r in regs[:10]])
# Candidate fixes: skip shadow_only when weak alignment (keep soft fragments)
m2 = steps[0][1].copy()
m2 = strip_transient_from_mask(m2, b, a)
m2 = strip_shadow_fragments_from_mask(m2, b, a, registration_ok=False)
m2 = strip_parking_cluster_from_mask(m2, b, a)
m2 = strip_weak_seasonal_veg_from_mask(m2, b, a)
m2 = recover_chromatic_roof_construction(m2, b_chr, a_chr)
m2 = recover_dark_roof_construction(m2, b_chr, a_chr)
# only soft fragment re-strip, no shadow_only
m2 = strip_shadow_fragments_from_mask(m2, b, a, registration_ok=False)
print("ALT_no_shadow_only px", count(m2), f"pct={100*count(m2)/m2.size:.3f}%")
regs2 = analyze_change_regions(
m2, a_chr, min_area=150, use_ensemble=False, before_img=b_chr,
registration_ok=False,
)
print("ALT_no_shadow_only regions", len(regs2), "top", [r["area"] for r in regs2[:8]])
# skip ALL shadow strips
m3 = steps[0][1].copy()
m3 = strip_transient_from_mask(m3, b, a)
m3 = strip_parking_cluster_from_mask(m3, b, a)
m3 = strip_weak_seasonal_veg_from_mask(m3, b, a)
m3 = recover_chromatic_roof_construction(m3, b_chr, a_chr)
m3 = recover_dark_roof_construction(m3, b_chr, a_chr)
print("ALT_no_shadow_at_all px", count(m3), f"pct={100*count(m3)/m3.size:.3f}%")
regs3 = analyze_change_regions(
m3, a_chr, min_area=150, use_ensemble=False, before_img=b_chr,
registration_ok=False,
)
print("ALT_no_shadow_at_all regions", len(regs3), "top", [r["area"] for r in regs3[:8]])
(out / "ablation.json").write_text(json.dumps({"ncc": ncc, "steps": rows}, indent=2), encoding="utf-8")
# Compare to run61 overlay / mask footprint if we can load overlay difference
print("run61 target ~193927 px (11.5%), run73 ~123141 (7.3%)")
if __name__ == "__main__":
main()
|