ZipArchive.tensor#

ZipArchive.tensor(device: str = 'cpu', normalize='imagenet', dtype: torch.dtype | None = None) → torch.Tensor#

Convert all images from this source into a single 4D PyTorch tensor.

Parameters:

  • device (str, optional) – target PyTorch device, e.g. "cpu", "cuda" or "mps", defaults to "cpu"

  • normalize (str, tuple, or None, optional) – normalisation to apply to each frame; passed directly to Image.tensor, defaults to "imagenet"

  • dtype (torch.dtype or None, optional) – output tensor dtype, for example torch.float32; passed directly to Image.tensor, defaults to None

Raises:

  • ImportError – if PyTorch is not installed

  • TypeError – if the source is not finite (no __len__)

  • TypeError – if any yielded item is not an Image

  • ValueError – if any frame has a different shape to the first

Returns:

tensor of shape (N, C, H, W)

Return type:

torch.Tensor

Images are decoded one at a time and written directly into a pre-allocated tensor, so peak memory is one decoded frame plus the output tensor — the full source need not reside in memory at once. This is particularly useful for VideoFile and ZipArchive.

Example:

>>> from machinevisiontoolbox import VideoFile
>>> t = VideoFile("traffic_sequence.mpg").tensor(normalize=None)
>>> t.shape
torch.Size([N, 3, H, W])