def main():

    tf.enable_eager_execution()

    vertices, faces = meshzoo.iso_sphere(2)
    faces = point_normals_outward(vertices, faces)

    vertices[:, 0] *= 0.5

    vertices = tf.constant(vertices, dtype=tf.float32)
    faces = tf.constant(faces, dtype=tf.int32)

    visualise_mesh(vertices, faces)

    offset_direction = tf.linalg.l2_normalize([-1, 0.05, 0.])
    offset_magnitudes = tf.where(tf.greater(vertices[:, 0], 0.45),
                                 tf.ones([len(vertices)]),
                                 tf.zeros([len(vertices)]))
    vertices = apply_offsets_with_pushing(vertices[None], faces,
                                          offset_direction[None],
                                          offset_magnitudes[None])[0]
    visualise_mesh(vertices, faces)

    offset_direction = tf.linalg.l2_normalize([0., 0.01, -1.])
    offset_magnitudes = tf.where(tf.greater(vertices[:, 2], 0.9),
                                 tf.ones([len(vertices)]),
                                 tf.zeros([len(vertices)]))
    vertices = apply_offsets_with_pushing(vertices[None], faces,
                                          offset_direction[None],
                                          offset_magnitudes[None])[0]
    visualise_mesh(vertices, faces)
Exemplo n.º 2
0
def write(filename, f):
    '''Write a function `f` defined in terms of spherical coordinates to a file.
    '''
    import meshio
    import meshzoo
    points, cells = meshzoo.iso_sphere(5)
    # get spherical coordinates from points
    polar = numpy.arccos(points[:, 2])
    azimuthal = numpy.arctan2(points[:, 1], points[:, 0])
    vals = f(polar, azimuthal)
    meshio.write(filename, points, {'triangle': cells}, point_data={'f': vals})
    return
Exemplo n.º 3
0
def create_sphere(n_subdivide=3, fix_normals=True):
    # 3 makes 642 verts, 1280 faces,
    # 4 makes 2562 verts, 5120 faces
    verts, faces = meshzoo.iso_sphere(n_subdivide)
    if fix_normals:
        v0 = verts[faces[:,0], :]
        v1 = verts[faces[:,1], :]
        v2 = verts[faces[:,2], :]
        v_m = (v0+v1+v2)/3
        e0 = v1-v0
        e1 = v2-v0
        p = np.cross(e0,e1)
        outwards = (v_m*p).sum(1)
        faces = np.where(outwards[:,None] > 0, faces[:,[0,1,2]], faces[:,[0,2,1]])
    return verts, faces
Exemplo n.º 4
0
        self.encoder = models.resnet18(pretrained=self.use_pretrained_encoder,
                                       num_classes=self.bottleneck_size)
        self.decoder = PointGenCon_SkipConection(
            bottleneck_size=self.bottleneck_size)

    def forward(self, x, grid):
        x = x[:, :3, :, :].contiguous()
        x = self.encoder(x)

        grid = grid.contiguous()

        y = self.decoder(x, grid)

        return y.transpose(2, 1).contiguous()


if __name__ == '__main__':
    import meshzoo

    x = torch.randn(8, 3, 256, 256).cuda()
    vert, face = meshzoo.iso_sphere(3)

    vert = torch.FloatTensor(vert).transpose(0,
                                             1)[None, :, :].repeat(8, 1,
                                                                   1).cuda()

    print(vert.shape)

    model = SVR_AtlasNet_SPHERE_Plus().cuda()

    print(model(x, vert).shape)
Exemplo n.º 5
0
def create_sphere(n_subdivide=3):
    # 3 makes 642 verts, 1280 faces,
    # 4 makes 2562 verts, 5120 faces
    verts, faces = meshzoo.iso_sphere(n_subdivide)
    return verts, faces
Exemplo n.º 6
0
    print("previous weight loaded")

    # a = sio.loadmat(os.path.join('data','points_sphere.mat'))
    # points_sphere = np.array(a['p'])
    # points_sphere = torch.FloatTensor(points_sphere).transpose(0,1).contiguous()
    # print(points_sphere.shape) # torch.Size([3, 7446])

    n_subdivide = 4

    # inputs of sr.Mesh class should be like following.
    # verts: (#batch, #vertices, 3)
    # faces: (#batch, #faces, 3)

    verts, faces = [
        elem.transpose(0, 1)[np.newaxis, ...]
        for elem in meshzoo.iso_sphere(n_subdivide)
    ]

    # print(verts.shape) # (1, 2562, 3)
    # print(faces.shape) # (1, 5120, 3)

    # logger
    test_loss = AverageValueMeter()
    test_curve = []

    test_loss.reset()
    model.eval()

    with torch.no_grad():
        for i, data in enumerate(tqdm.tqdm(dataloader), 0):
            img, points, label, _, _ = data
Exemplo n.º 7
0
        self.avg = 0
        self.sum = 0
        self.count = 0.0

    def update(self, val, n=1):
        self.val = val
        self.sum += val * n
        self.count += n
        self.avg = self.sum / self.count


if __name__ == "__main__":
    import numpy as np
    import meshzoo

    verts, faces = meshzoo.iso_sphere(3)
    verts = verts[np.newaxis, ...]
    faces = faces[np.newaxis, ...]

    output_path = 'logs/test_gif_01.gif'
    video = save_as_gif(verts,
                        faces,
                        output_path,
                        input_img=None,
                        verbose=True)
    print(video.shape)

    output_path = 'logs/test_gif_02'
    save_as_gif(verts, faces, output_path, input_img=None, verbose=False)
    save_as_obj(verts, faces, output_path, verbose=True)
Exemplo n.º 8
0
def test_iso_sphere():
    points, cells = meshzoo.iso_sphere()
    assert len(points) == 2562
    assert _near_equal(numpy.sum(points, axis=0), [0.0, 0.0, 0.0])
    assert len(cells) == 5120
    return