vis_hit_sample.py

import argparse

from datasets import load_dataset

import open3d as o3d
import pyvista as pv
from PIL import Image
import matplotlib
import matplotlib.pyplot as plt

import numpy as np
import random


def plot_3D_image(values, resolution, p=None, interactive_slice=False, orthogonal_slices=True):
    ''' Interactive plot of the 3D volume'''

    # Create the spatial reference
    grid = pv.ImageData()
    values = np.transpose(values, (1,2,0))
    # Set the grid dimensions: shape + 1 because we want to inject our values on
    # the CELL data
    grid.dimensions = np.array(values.shape) + 1

    # Edit the spatial reference
    # The bottom left corner of the data set
    origin = np.array(resolution[0]) * np.array(values.shape) * 0.5
    grid.origin = origin
    #print(f'Scan size in meter: {origin * 2}')
    grid.spacing = resolution[0]  # These are the cell sizes along each axis

    # Add the data values to the cell data
    grid.cell_data["values"] = values.flatten(order="F")  # Flatten the array!

    if p is None:
        p = pv.Plotter()

    if orthogonal_slices:
        slices = grid.slice_orthogonal()
        cmap = matplotlib.colors.ListedColormap(['black', 'indianred', 'goldenrod', 'steelblue', 'ghostwhite'])
        p.add_mesh(slices, cmap=cmap)

    if interactive_slice:
        p.add_mesh_clip_plane(grid)

    return p


def get_sliced_mri_png(sample):
    # get data
    mri = np.asarray(sample['mri_seg'])
    resolution = np.asarray(sample['resolution'])
    # set plotter
    p = pv.Plotter(shape=(1, 1), off_screen=True)
    p.subplot(0, 0)
    plotter = plot_3D_image(mri, resolution, p, interactive_slice=False, orthogonal_slices=True)
    plotter.view_yz()
    plotter.remove_scalar_bar()
    # store screenshot
    img = p.screenshot("./extras/img.png", return_img=True)
    # read screenshot
    img = Image.fromarray(img)
    # set plotter lateral
    p = pv.Plotter(shape=(1, 1), off_screen=True)
    p.subplot(0, 0)
    plotter = plot_3D_image(mri, resolution, p, interactive_slice=False, orthogonal_slices=True)
    plotter.remove_scalar_bar()
    plotter.view_xz()
    img_lateral = p.screenshot("./extras/img_lateral.png", return_img=True)
    img_lateral = Image.fromarray(img_lateral)
    # resize
    img = img.resize((512+128, 372+128))
    img_lateral = img_lateral.resize((512+128, 372+128))
    return img, img_lateral


def vis_hit_sample(sample):
    """
    :param sample: HIT dataset sample
    :return:
    """
    # get point-cloud from sample
    pc = np.asarray(sample['body_cont_pc'])
    # get mesh and mesh-free-verts from sample
    mesh_verts = np.asarray(sample['smpl_dict']['verts'])
    mesh_verts_free = np.asarray(sample['smpl_dict']['verts_free'])
    mesh_faces = np.asarray(sample['smpl_dict']['faces'])

    # create point-cloud
    pcd = o3d.geometry.PointCloud()
    pcd.points = o3d.utility.Vector3dVector(pc)
    pcd.paint_uniform_color([0.6509803922, 0.2901960784, 0.2823529412])
    pcd_front = pcd.__copy__()

    # create mesh
    mesh = o3d.geometry.TriangleMesh()
    mesh.vertices = o3d.utility.Vector3dVector(mesh_verts)
    mesh.triangles = o3d.utility.Vector3iVector(mesh_faces)
    mesh.paint_uniform_color([0.737254902, 0.7960784314, 0.8196078431])

    # create mesh-free-verts
    mesh_free = o3d.geometry.TriangleMesh()
    mesh_free.vertices = o3d.utility.Vector3dVector(mesh_verts_free)
    mesh_free.triangles = o3d.utility.Vector3iVector(mesh_faces)
    mesh_free.paint_uniform_color([0.737254902, 0.7960784314, 0.8196078431])

    # visualize sample
    vis = o3d.visualization.Visualizer()
    vis.create_window()
    # side-view
    xyz = (-np.pi / 2, 0, 0)
    R1 = o3d.geometry.get_rotation_matrix_from_xyz(xyz)
    # vis mesh with pointcloud
    vis.add_geometry(mesh.rotate(R1, center=(0, 0, 0)))
    vis.add_geometry(pcd.rotate(R1, center=(0, 0, 0)))
    # vis mesh-free-verts with pointcloud
    vis.add_geometry(mesh_free.translate((1.2, 0, 0)))
    vis.add_geometry(mesh_free.rotate(R1, center=(0, 0, 0)))
    vis.add_geometry(pcd_front.translate((1.2, 0, 0)))
    vis.add_geometry(pcd_front.rotate(R1, center=(0, 0, 0)))
    # render
    vis.get_render_option().mesh_show_wireframe = True
    vis.get_render_option().point_size = 2
    vis.poll_events()
    vis.update_renderer()
    vis.run()
    return 0


if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='HIT dataset visualization')
    parser.add_argument('--gender', type=str, default='male')
    parser.add_argument('--split', type=str, default='train')
    parser.add_argument('--idx', type=int, default=None)
    parser.add_argument('--show_skin', action='store_true')
    parser.add_argument('--show_tissue', action='store_true')

    # get args
    args = parser.parse_args()
    assert args.gender in ['male', 'female']
    assert args.split in ['train', 'validation', 'test']

    # load split
    hit_dataset = load_dataset("varora/hit", name=args.gender, split=args.split)

    # to load specific split, use:
    # male splits
    #male_val = load_dataset("varora/hit", "male", split="validation")
    #male_val = load_dataset("varora/hit", "male", split="validation")
    #male_test = load_dataset("varora/hit", "male", split="test")
    # female splits
    #female_train = load_dataset("varora/hit", "female", split="train")
    #female_val = load_dataset("varora/hit", "female", split="validation")
    #female_test = load_dataset("varora/hit", "female", split="test")

    # len of split
    N_dataset = hit_dataset.__len__()

    # get idx for sample
    if not args.idx:
        idx = random.randint(0, N_dataset)
    else:
        idx = args.idx
        assert idx < N_dataset, f"{idx} in {args.gender}:{args.split} is out of range for dataset of length {N_dataset}."

    # get sample
    hit_sample = hit_dataset[idx]
    # visualize the sample
    print(f"Visualizing sample no. {idx} in {args.gender}:{args.split}.")
    if args.show_tissue:
        img, img_lateral = get_sliced_mri_png(hit_sample)
        img.show()
        img_lateral.show()
    elif args.show_skin:
        vis_hit_sample(hit_sample)
    else:
        img, img_lateral = get_sliced_mri_png(hit_sample)
        img.show()
        img_lateral.show()
        vis_hit_sample(hit_sample)