import numpy as np

META = {
    "name": "euler_morphology_denoising",
    "domain": "vision",
    "description": "Synthetic binary-shape image denoising with spatial topology in the clean target."
}

def get_dataset(seed, n_train, n_test):
    rng = np.random.RandomState(seed)
    h = w = 16
    yy, xx = np.mgrid[:h, :w]

    def make(n):
        clean = np.zeros((n, 1, h, w), dtype=np.float32)
        for i in range(n):
            if rng.rand() < 0.5:
                cx, cy = rng.uniform(3, 13, 2)
                r = rng.uniform(2.0, 4.5)
                clean[i, 0] = (((xx-cx)**2 + (yy-cy)**2) <= r*r)
            else:
                x0, y0 = rng.randint(1, 10, 2)
                bw, bh = rng.randint(3, 7, 2)
                clean[i, 0, y0:y0+bh, x0:x0+bw] = 1.0
            if rng.rand() < 0.25:
                cx, cy = rng.uniform(3, 13, 2)
                r = rng.uniform(1.2, 2.5)
                blob = (((xx-cx)**2 + (yy-cy)**2) <= r*r)
                clean[i, 0] = np.maximum(clean[i, 0], blob)
        noisy = np.clip(clean + rng.normal(0, 0.28, clean.shape), 0, 1)
        return noisy.astype(np.float32), clean

    xtr, ytr = make(n_train)
    xte, yte = make(n_test)
    return {"xtr": xtr, "ytr": ytr, "xte": xte, "yte": yte,
            "task": "regression", "metric": "mse", "out_dim": 1}