import numpy as np

def select_mask(masks, rule="largest"):
    if rule == "largest":
        return max(masks, key=lambda m: m["area"])
    raise ValueError("Unknown mask-selection rule")

def pixel_to_metric(mask, depth, fx_px, fy_px):
    import cv2
    
    # Handle shape mismatch between mask and depth
    if mask.shape != depth.shape:
        print(f"⚠️  Shape mismatch: mask {mask.shape} vs depth {depth.shape}")
        # Resize mask to match depth dimensions
        if mask.dtype == bool:
            mask_uint8 = mask.astype(np.uint8) * 255
        else:
            mask_uint8 = mask
        mask_resized = cv2.resize(mask_uint8, (depth.shape[1], depth.shape[0]), 
                                interpolation=cv2.INTER_NEAREST)
        mask = mask_resized > 127  # Convert back to boolean
        print(f"✅ Resized mask to {mask.shape}")
    
    ys, xs = np.where(mask)
    if len(xs) == 0 or len(ys) == 0:
        print("⚠️  Empty mask - no object pixels found")
        return dict(distance_m=0.0, width_m=0.0, height_m=0.0)
        
    bbox_w, bbox_h = xs.max() - xs.min(), ys.max() - ys.min()
    D = float(np.median(depth[mask]))
    width_m  = (bbox_w * D) / fx_px
    height_m = (bbox_h * D) / fy_px
    return dict(distance_m=D, width_m=width_m, height_m=height_m)