Utilities

cvcuda.as_image(buffer: cvcuda.ExternalBuffer, format: cvcuda.Format = nvcv.Format.NONE) → cvcuda.Image
cvcuda.as_image(buffer: collections.abc.Sequence[object] = [], format: cvcuda.Format = nvcv.Format.NONE) → cvcuda.Image

Overloaded function.

  1. as_image(buffer: cvcuda.ExternalBuffer, format: cvcuda.Format = nvcv.Format.NONE) -> cvcuda.Image

Wrap an external buffer as an image and tie the buffer lifetime to the image

  1. as_image(buffer: collections.abc.Sequence[object] = [], format: cvcuda.Format = nvcv.Format.NONE) -> cvcuda.Image

Wrap a vector of external buffers as an image and tie the buffer lifetime to the image

cvcuda.as_images(buffers: collections.abc.Sequence[object] = [], format: cvcuda.Format = nvcv.Format.NONE) → cvcuda.ImageBatchVarShape

Wrap a vector of external buffers as a batch of images, and tie the buffers lifetime to it

cvcuda.as_stream(arg: torch.cuda.Stream, /) → cvcuda.Stream
cvcuda.as_stream(arg: cupy.cuda.Stream, /) → cvcuda.Stream
cvcuda.as_stream(arg: ctypes.c_void_p, /) → cvcuda.Stream
cvcuda.as_stream(arg: int, /) → cvcuda.Stream

Note

Pass an owning torch.cuda.Stream or cupy.cuda.Stream object when possible. CV-CUDA retains that owner while the wrapped stream or a resource carrying its stream dependency is in use. Raw CUDA stream handles and ExternalStream objects remain caller-owned; their underlying stream must stay valid for all CV-CUDA uses, including while resources carry dependencies on it.

cvcuda.as_tensor(buffer: cvcuda.ExternalBuffer, layout: cvcuda.TensorLayout | str | None = None) → cvcuda.Tensor
cvcuda.as_tensor(image: cvcuda.Image) → cvcuda.Tensor

Overloaded function.

  1. as_tensor(buffer: cvcuda.ExternalBuffer, layout: cvcuda.TensorLayout | str | None = None) -> cvcuda.Tensor

Wrap an existing buffer into a Tensor object with the given layout.

  1. as_tensor(image: cvcuda.Image) -> cvcuda.Tensor

Wrap an existing image into a Tensor object.

cvcuda.as_tensors(buffers: collections.abc.Sequence[object] = [], layout: cvcuda.TensorLayout | None = None) → cvcuda.TensorBatch

Wrap a list of external buffers as a batch of tensors, and tie the buffers lifetime to it

cvcuda.reshape(tensor: cvcuda.Tensor, shape: tuple, layout: cvcuda.TensorLayout | str | None = None) → cvcuda.Tensor

Produces a tensor pointing to the same data but with a new shape and layout.