# Fit the user's front screenshot (Blender Front Orthographic: screen right = model +X) to the model:
#   u = a * x + bu,  v = -a * z + bv
# from the orange selection outline's lower edge (the hull's bottom silhouette), then convert the red strokes.
import json
import numpy as np
from PIL import Image
from collections import deque
S = r"C:\Users\PC\AppData\Local\Temp\claude\D--HADAL---The-Descent\1fbd6834-d7f7-4606-97c8-f13f3d7e4dae"
F = json.load(open(S + r"\scratchpad\front_tables.json"))
low = np.array(F["low"])
im = np.asarray(Image.open(S + r"\images\4.png").convert("RGB")).astype(int)
H, W, _ = im.shape
R, G, B = im[..., 0], im[..., 1], im[..., 2]
org = (R > 200) & (G > 110) & (G < 190) & (B < 80)
red = (R > 170) & (G < 90) & (B < 90)
# lower outline: for each column, the lowest orange pixel (the outline runs 2-3 px outside the silhouette)
bot = {}
for u in range(380, 1320, 3):
    idx = np.nonzero(org[:, u])[0]
    if len(idx):
        bot[u] = idx[-1] - 2.0
def cost(p):
    a, bu, bv = p
    r = []
    for u, v in bot.items():
        x = (u - bu) / a
        if -1.05 < x < 1.05:
            r.append(v - (-a * np.interp(x, low[:, 0], low[:, 1]) + bv))
    r = np.array(r)
    return float(np.mean(np.minimum(r * r, 225.0))), r
p = np.array([550.0, 850.0, 435.0])
best = (cost(p)[0], p.copy()); step = np.array([20.0, 20.0, 20.0])
for it in range(300):
    imp = False
    for k in range(3):
        for s in (1, -1):
            q = best[1].copy(); q[k] += s * step[k]
            c = cost(q)[0]
            if c < best[0]:
                best = (c, q); imp = True
    if not imp:
        step *= 0.5
        if step.max() < 0.01:
            break
a, bu, bv = best[1]
c, r = cost(best[1])
print("front fit a=%.2f px/m bu=%.1f bv=%.1f rms=%.2f px median=%.2f n=%d" % (a, bu, bv, np.sqrt(c), np.median(np.abs(r)), len(r)))
# red strokes, components (drop the thin X-axis line)
lab = np.zeros((H, W), int); comps = []
n = 0
for v0, u0 in zip(*np.nonzero(red)):
    if lab[v0, u0]:
        continue
    n += 1; q = deque([(v0, u0)]); lab[v0, u0] = n; pts = []
    while q:
        v, u = q.popleft(); pts.append((u, v))
        for dv in (-2, -1, 0, 1, 2):
            for du in (-2, -1, 0, 1, 2):
                vv, uu = v + dv, u + du
                if 0 <= vv < H and 0 <= uu < W and red[vv, uu] and not lab[vv, uu]:
                    lab[vv, uu] = n; q.append((vv, uu))
    comps.append(pts)
comps = [c for c in comps if len(c) > 1000]
comps.sort(key=len, reverse=True)
out = {"a": a, "bu": bu, "bv": bv, "strokes": []}
for c in comps:
    out["strokes"].append([((u - bu) / a, (bv - v) / a) for u, v in c])
json.dump(out, open(S + r"\scratchpad\front_fit.json", "w"))
# check picture: the model's bottom silhouette in blue
img = Image.open(S + r"\images\4.png").convert("RGB"); px = img.load()
for x, z in low:
    if z > 5:
        continue
    uu, vv = int(round(a * x + bu)), int(round(-a * z + bv))
    if 0 <= uu < W and 0 <= vv < H:
        for d in (-1, 0, 1):
            px[uu, min(H - 1, max(0, vv + d))] = (40, 160, 255)
img.save(S + r"\scratchpad\front_fit_check.png")
for i, c in enumerate(out["strokes"]):
    a_ = np.array(c)
    print("stroke", i, len(c), "x %.3f..%.3f  z %.3f..%.3f" % (a_[:, 0].min(), a_[:, 0].max(), a_[:, 1].min(), a_[:, 1].max()))