DiffAtlas-2shot/Dataset/sdf.py

import numpy as np
from scipy.ndimage import distance_transform_edt, gaussian_filter, zoom, median_filter
import matplotlib.pyplot as plt
from skimage.measure import marching_cubes
from scipy.ndimage import binary_fill_holes, label
from mpl_toolkits.mplot3d.art3d import Poly3DCollection
from scipy.spatial import Delaunay
import open3d as o3d
import pyvista as pv
from scipy.interpolate import griddata
# import trimesh

def compute_sdf(voxel_matrix):
    inside_dist = distance_transform_edt(voxel_matrix)
    outside_dist = distance_transform_edt(1 - voxel_matrix)
    sdf = outside_dist - inside_dist
    # sdf = gaussian_filter(sdf, sigma=3.0)
    sdf = sdf / 32.0
    sdf = sdf.clip(-0.2, 0.2)
    return sdf

def visualize_sdf(voxel_matrix, sdf, slice_index, prefix='slice'):
    plt.figure(figsize=(12, 6))
    plt.subplot(1, 2, 1)
    plt.title("Voxel Matrix Slice")
    plt.imshow(voxel_matrix[slice_index], cmap='gray')
    plt.subplot(1, 2, 2)
    plt.title("SDF Slice")
    plt.imshow(sdf[slice_index], cmap='jet')
    plt.colorbar()
    plt.show()
    plt.savefig(f'{prefix}_{slice_index}.png')
    plt.close()


def extract_surface(sdf):
    vertices, faces, normals, values = marching_cubes(sdf, level=0)
    return vertices, faces


def visualize_surface(vertices, faces):
    fig = plt.figure()
    ax = fig.add_subplot(111, projection='3d')
    mesh = Poly3DCollection(vertices[faces])
    mesh.set_edgecolor('k')
    ax.add_collection3d(mesh)
    ax.set_xlim(0, 32)
    ax.set_ylim(0, 32)
    ax.set_zlim(0, 32)
    plt.show()
    plt.savefig('surface.png')
    plt.close()


def create_3d_mask(vertices, faces, shape):
    mask = np.zeros(shape, dtype=np.uint8)
    
    # Create a Delaunay triangulation for the vertices
    delaunay = Delaunay(vertices)
    
    # Generate grid points
    x = np.arange(0, shape[0])
    y = np.arange(0, shape[1])
    z = np.arange(0, shape[2])
    grid = np.array(np.meshgrid(x, y, z)).T.reshape(-1, 3)
    
    # Find grid points that are inside the convex hull
    inside = delaunay.find_simplex(grid) >= 0
    
    # Convert the grid points back to 3D index
    grid_points = grid[inside]
    
    # Set the mask values to 1 where points are inside the convex hull
    mask[grid_points[:, 0], grid_points[:, 1], grid_points[:, 2]] = 1
    
    return mask


def fill_holes_along_axis(volume, axis):
    filled_volume = np.copy(volume)
    for i in range(volume.shape[axis]):
        if axis == 0:
            slice_2d = volume[i, :, :]
        elif axis == 1:
            slice_2d = volume[:, i, :]
        else:
            slice_2d = volume[:, :, i]

        filled_slice = binary_fill_holes(slice_2d)

        # 确保单连通性
        labeled_array, num_features = label(filled_slice)
        if num_features > 1:
            max_label = max(range(1, num_features + 1), key=lambda x: np.sum(labeled_array == x))
            filled_slice = labeled_array == max_label

        if axis == 0:
            filled_volume[i, :, :] = filled_slice
        elif axis == 1:
            filled_volume[:, i, :] = filled_slice
        else:
            filled_volume[:, :, i] = filled_slice

    return filled_volume


def mesh_to_voxel(vertices, faces):

    shape = [64, 64, 64]
    faces = np.hstack([[3] + list(face) for face in faces])
    mesh = pv.PolyData(vertices, faces)
    grid = pv.voxelize(mesh, density=1.0)
    voxel_points = np.array(grid.points)
    filled_voxel = np.zeros(shape)

    for voxel in voxel_points:
        x, y, z = voxel
        x, y, z = int(np.round(x)), int(np.round(y)), int(np.round(z))  # 将顶点坐标转换为整数索引
        if 0 <= x < shape[0] and 0 <= y < shape[1] and 0 <= z < shape[2]:
            filled_voxel[x, y, z] = 1  # 设置为1表示该位置有网格顶点
    return filled_voxel


def sdf_to_filled_mesh(sdf, level=0.02):
    # vertices, faces, _, _ = marching_cubes(sdf, level=level)
    # filled_voxel = mesh_to_voxel(vertices, faces)

    # for _ in range(1):
    #     filled_voxel = fill_holes_along_axis(filled_voxel, 0)
    #     filled_voxel = fill_holes_along_axis(filled_voxel, 1)
    #     filled_voxel = fill_holes_along_axis(filled_voxel, 2)
    
    # while True:
    #     prev_filled_voxel = filled_voxel.copy()
    #     filled_voxel = fill_holes_along_axis(filled_voxel, 0)
    #     filled_voxel = fill_holes_along_axis(filled_voxel, 1)
    #     filled_voxel = fill_holes_along_axis(filled_voxel, 2)

    #     if np.array_equal(filled_voxel, prev_filled_voxel):
    #         break

    vertices_filled, faces_filled, normals_filled, values_filled = marching_cubes(sdf, level=level)

    # 将网格转换为 Open3D 的三角网格对象
    mesh = o3d.geometry.TriangleMesh()
    mesh.vertices = o3d.utility.Vector3dVector(vertices_filled)
    mesh.triangles = o3d.utility.Vector3iVector(faces_filled)
    mesh.compute_vertex_normals()

    # 拉普拉斯平滑
    mesh = mesh.filter_smooth_simple(number_of_iterations=20)
    mesh.compute_vertex_normals()

    return np.asarray(mesh.vertices), np.asarray(mesh.triangles)


def sdf_to_voxel(sdf, level=0.02):

    sdf_smoothed = gaussian_filter(sdf, sigma=0.5)
    binary_mask = binary_fill_holes(sdf_smoothed <= level).astype(np.uint8)
    # structuring_element = ball(radius=1)
    # binary_mask_opened = binary_opening(binary_mask, footprint=structuring_element)
    # binary_mask_closed = binary_closing(binary_mask_opened, footprint=structuring_element)
    # binary_mask_eroded = binary_erosion(binary_mask_closed, structure=structuring_element)
    # binary_mask_dilated = binary_dilation(binary_mask_eroded, structure=structuring_element)
    # filled_voxel = gaussian_filter(binary_mask_dilated.astype(np.float32), sigma=1.0)
    # filled_voxel = filled_voxel > 0.5
    filled_voxel = binary_mask
    
    return filled_voxel