demo code for visualizing the prediction on 3D object detection and outdoor segmentation tasks about ogc

[amiltonwong]

@Szy-Young , Thanks a lot, I'll try it.

[Szy-Young]

Hi @amiltonwong

You can use the following code for a quick visualization with Open3D:

# pc: (N, 3) numpy array
# segm: (N,) numpy array

import open3d as o3d
from utils.visual_util import build_pointcloud

pcds = [build_pointcloud(pc, segm)]
o3d.visualization.draw_geometries(pcds)

For the visualization effects in our paper and video demo, you can first use the following code to convert point clouds into meshes of spheres:

# pc: (N, 3) numpy array
# segm: (N,) numpy array

import open3d as o3d

def pc_segm_to_sphere(pc, segm=None, radius = 0.01, resolution=10, with_background=False, default_color=COLORGRAY2):
    """
    Visualize point cloud as mesh balls. The color denotes hard segmentation.
    :param pc: (N, 3)
    :param segm: (N,)
    """
    assert pc.shape[1] == 3 and len(pc.shape) == 2, f"Point cloud is of size {pc.shape} and cannot be displayed!"
    n_point = pc.shape[0]
    if with_background:
        colors = np.concatenate((COLOR20[-1:], COLOR20[:-1]), axis=0)
    else:
        colors = COLOR20

    meshes = []
    for pid in range(n_point):
        mesh = o3d.geometry.TriangleMesh.create_sphere(radius=radius, resolution=resolution)
        if segm is not None:
            mesh.paint_uniform_color(colors[segm[pid] % colors.shape[0]] / 255.)
        else:
            mesh.paint_uniform_color(default_color / 255.)
        mesh.translate(pc[pid])
        meshes.append(mesh)

    # Merge
    mesh = meshes[0]
    for i in range(1, len(meshes)):
        mesh += meshes[i]
    return mesh

mesh = pc_segm_to_sphere(pc, segm)
save_file = ...
o3d.io.write_triangle_mesh(save_file, mesh)

Then you can render the mesh models with softwares like KeyShot (our choice), Blender, etc.