Return the tokens from `x` corresponding to the indices in `idx`.

Reshape a set of token vectors into an image. This is the reverse operation of `image_as_tokens`.

Reshape an image of shape (b, c, h, w) to a set of vectorized patches of shape (b, h*w/p^2, c*p^2). In other words, the set of non-overlapping

Create sine-cosine positional embeddings. Args: n: the number of embedding vectors, corresponding to the number of tokens (patche

Create a randomly permuted token-index tensor for determining which tokens to mask. Args: bs: batch size n_tok: numbe

Mask the image patches uniformly at random, as described in the paper: the patch tokens are randomly permuted (per image), and the first N are

Return a copy of the image batch, with the masked patches set to 0. Used for visualization.

When True, the previously computed mask will be used on new inputs, instead of creating a new one.