spdl.io.async_load_image_batch_nvdec

async async_load_image_batch_nvdec(srcs: list[str | bytes], *, device_config: CUDAConfig, width: int | None, height: int | None, pix_fmt: str | None = 'rgba', demux_config: DemuxConfig | None = None, decode_options: dict[str, Any] | None = None, strict: bool = True, **kwargs) → CUDABuffer[source]

[Experimental] Batch load images with NVDEC.

This function combines demux_image() and decode_packets_nvdec() to produce buffer object from source in one step.

It concurrently demuxes the input images, using the ThreadPoolExecutor attached to the running async event loop, fetched by get_running_loop().

Parameters:

  • srcs – List of source identifiers.

  • device_config – The CUDA device config. See transfer_buffer().

  • width – Optional: Resize the frame.

  • height – Optional: Resize the frame.

  • pix_fmt – Optional: Change the format of the pixel.

  • demux_config – Optional: Demux options passed to async_demux_media().

  • decode_options – Optional: Other decode options passed to async_decode_packets_nvdec().

  • strict – Optional: If True, raise an error if any of the images failed to load.

Returns:

A buffer object.

Example

>>> srcs = [
...     "sample1.jpg",
...     "sample2.png",
... ]
>>> coro = async_load_image_batch_nvdec(
...     srcs,
...     cuda_device_index=0,
...     width=124,
...     height=96,
...     pix_fmt="rgba",
... )
>>> buffer = asyncio.run(coro)
>>> array = spdl.io.to_torch(buffer)
>>> # An array with shape NCHW==[2, 4, 96, 124] on CUDA device 0
>>>