ImageDataGenerator for 3D images

[josejimenezluna]

Hi,

Does the ImageDataGenerator work on 3d images (5 dimensional array (n_samples, channels, x, y, z)) as well? I looked at the source code and I'm not sure if this is the case.

def random_rotation(x, rg, row_index=1, col_index=2, channel_index=0,
                    fill_mode='nearest', cval=0.):
    theta = np.pi / 180 * np.random.uniform(-rg, rg)
    rotation_matrix = np.array([[np.cos(theta), -np.sin(theta), 0],
                                [np.sin(theta), np.cos(theta), 0],
                                [0, 0, 1]])

    h, w = x.shape[row_index], x.shape[col_index]
    transform_matrix = transform_matrix_offset_center(rotation_matrix, h, w)
    x = apply_transform(x, transform_matrix, channel_index, fill_mode, cval)
    return x

[joelthchao]

ImageDataGenerator doesn't work on 3d images. It only takes 4d array as input.

[fchollet]

But it would be reasonably easy to reimplement for 3D data a subset of the operations done by ImageDataGenerator.

[dhuy228]

@Hawk31 hey mate, I just customised the Keras' ImageDataGenerator for medical image analysis applications, i.e. 3D Dicom slices (10,32,32,32,1). It might help with your problem.

https://github.com/dhuy228/augmented-volumetric-image-generator/blob/master/generator.py

[Mcitrin]

@dhuy228

could you give an example how how your calling your custom ImageDataGenerator

im trying to do a similar thing using .nii files generated from a number of dicom slices

the shape of the array of patches in (1000,32,32,32) (num patches, x, y, z) and i re shaped it to be (1000, 32, 32, 32, 1) the script seems augmenting my image array properly but it dosn't augment the labels accordingly

im nt sure what i need to do to fix this