# Wizualizacja mapy glebi DAv2-Small na realnych ujeciach + statystyki
# wglebienia wnetrza skrzyni vs plaszczyzny przednie.
import numpy as np
import torch
from PIL import Image

from transformers import AutoImageProcessor, AutoModelForDepthEstimation

MODEL_ID = "depth-anything/Depth-Anything-V2-Small-hf"
CASES = [
    (r"C:\xampp\htdocs\vilmax\vilmal\tmp\recolor-svc-storage-granat-v4.png", "storage"),
    (r"C:\xampp\htdocs\vilmax\vilmax-cockpit\data\hall-audit\preview\p-6419.jpg", "front"),
]

proc = AutoImageProcessor.from_pretrained(MODEL_ID)
model = AutoModelForDepthEstimation.from_pretrained(MODEL_ID).to("cuda").eval()

for path, tag in CASES:
    img = Image.open(path).convert("RGB")
    inputs = proc(images=img, return_tensors="pt").to("cuda")
    with torch.no_grad():
        depth = model(**inputs).predicted_depth[0]  # wieksze = BLIZEJ (dysparytancja)
    d = depth.cpu().numpy()
    d = (d - d.min()) / max(1e-6, d.max() - d.min())  # 0..1, 1=blisko
    Image.fromarray((d * 255).astype(np.uint8)).resize(img.size, Image.BILINEAR).save(
        rf"C:\xampp\htdocs\vilmax\vilmal\tmp\depth-{tag}.png"
    )
    print(tag, "img", img.size, "depth", d.shape, "min/max", round(float(d.min()),3), round(float(d.max()),3))