# Skan progow deficytu + czy kotwica 'plate' trafia w komponent komory.
import cv2
import numpy as np
import torch
from PIL import Image
from transformers import AutoImageProcessor, AutoModelForDepthEstimation

MODEL_ID = "depth-anything/Depth-Anything-V2-Small-hf"
IMG = r"C:\xampp\htdocs\vilmax\vilmal\tmp\recolor-svc-storage-granat-v4.png"
MATTE = r"C:\xampp\htdocs\vilmax\vilmal\tmp\birefnet-matte-6413.png"

proc = AutoImageProcessor.from_pretrained(MODEL_ID)
model = AutoModelForDepthEstimation.from_pretrained(MODEL_ID).to("cuda").eval()
img = Image.open(IMG).convert("RGB")
W, H = img.size
inputs = proc(images=img, return_tensors="pt").to("cuda")
with torch.no_grad():
    d = model(**inputs).predicted_depth[0].cpu().numpy()
d = cv2.resize(d, (W, H), interpolation=cv2.INTER_LINEAR)
d = (d - d.min()) / max(1e-6, d.max() - d.min())

matte = np.asarray(Image.open(MATTE).convert("L").resize((W, H), Image.BILINEAR), np.float32) / 255.0
mask = (matte > 0.5).astype(np.uint8)

for R in (200, 300, 400):
    k = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (2 * R + 1, 2 * R + 1))
    local_max = cv2.dilate(d * mask, k)
    deficit = np.where(mask > 0, local_max - d, 0).astype(np.float32)
    for th in (0.15, 0.2, 0.25, 0.3, 0.35):
        rec = (deficit > th).astype(np.uint8)
        n, lab, stats, _ = cv2.connectedComponentsWithStats(rec, 8)
        big = sorted(
            (int(s[cv2.CC_STAT_AREA]), int(s[cv2.CC_STAT_LEFT]), int(s[cv2.CC_STAT_TOP]),
             int(s[cv2.CC_STAT_WIDTH]), int(s[cv2.CC_STAT_HEIGHT]))
            for s in stats[1:] if s[cv2.CC_STAT_AREA] > 8000)
        big.sort(reverse=True)
        print(f"R={R} th={th}: comps>8k: {big[:6]}")