while looking through some deep learning implementation code, I found some strange list indexing in python code. Here is an example.

        if reference_points.shape[-1] == 2:
            offset_normalizer = torch.stack([input_spatial_shapes[..., 1], input_spatial_shapes[..., 0]], -1)
            sampling_locations = reference_points[:, :, None, :, None, :] \
                                 + sampling_offsets / offset_normalizer[None, None, None, :,

It was using None as an index when indexing a python object. First I thought is this indexing None on a python list and tested it. But it gave me an error.

It turns out that None indexing does work under numpy arrays and also tensor objects. Here is an example showing how none indexing with a numy array works.

import numpy as np

a = np.random.rand(3,2)


print(a[None,:].shape) # output: (1, 3, 2)
print(a[:,None].shape) # output: (3, 1, 2)
print(a[:,None,:,None].shape) # output: (3, 1, 2, 1)
print(a[:,None,None, :].shape) # output: (3, 1, 1, 2)

print(a[:,None,:,:]) # IndexError: too many indices for array: array is 2-dimensional, but 3 were indexed

As you can see, it practically acts as np.expand_dims function to add a dummy axis where I want, but in a more inline coding fashion.


1 Comment

arta · August 30, 2022 at 12:09 am

None in here stands for “Not only next end,” and it will return the rest of the rest of the elements in the list.

Leave a Reply

Your email address will not be published.