spdl.io.encode_image

encode_image(path: str, data: Array, pix_fmt: str = 'rgb24', **kwargs)

Save the given image array/tensor to file.

Parameters:

• path – The path to which the data are written.

• data (NumPy NDArray, PyTorch Tensor) –

  Image data in array format. The data must be uint8 type, either on CPU or CUDA device.

  The shape must be one of the following and must match the value of pix_fmt.

  • (height, width, channel==3) when pix_fmt="rgb24"

  • (height, width) when pix_fmt=gray8

  • (channel==3, height, width) when pix_fmt="yuv444p"

• pix_fmt – See above.

• encode_config (EncodeConfig) – Customize the encoding.

Example - Save image as PNG with resizing

>>> import asyncio
>>> import numpy as np
>>> import spdl.io
>>>
>>> data = np.random.randint(255, size=(32, 16, 3), dtype=np.uint8)
>>> coro = spdl.io.async_encode_image(
...     "foo.png",
...     data,
...     pix_fmt="rgb24",
...     encode_config=spdl.io.encode_config(
...         width=198,
...         height=96,
...         scale_algo="neighbor",
...     ),
... )
>>> asyncio.run(coro)
>>>

Example - Save CUDA tensor as image

>>> import torch
>>>
>>> data = torch.randint(255, size=(32, 16, 3), dtype=torch.uint8, device="cuda")
>>>
>>> async def encode(data):
>>>     buffer = await spdl.io.async_transfer_buffer_cpu(data)
>>>     return await spdl.io.async_encode_image(
...         "foo.png",
...         buffer,
...         pix_fmt="rgb24",
...     )
...
>>> asyncio.run(encode(data))
>>>