mindscience.common.unpatchify

mindscience.common.unpatchify(labels, img_size=(192, 384), patch_size=16, nchw=False)[source]

Convert a sequence of flattened patches back into an image-like tensor.

Parameters

  • labels (Tensor) – A tensor containing flattened patch representations. The shape of labels is expected to be \((N, num_patches, patch_size * patch_size * C)\), where C denotes the number of output channels.

  • img_size (tuple(int), optional) – Input image size. Default (192, 384).

  • patch_size (int, optional) – The patch size of image. Default 16.

  • nchw (bool, optional) – If True, the unpatchify shape contains \((N, C, H, W)\); if False, the unpatchify shape contains \((N, H, W, C)\). Default: False.

Returns

Tensor, with shape of \((N, H, W, C)\) if nchw is False or \((N, C, H, W)\) if nchw is True.