nvidia.dali.fn.experimental.decoders.image_slice

nvidia.dali.fn.experimental.decoders.image_slice(__data, __anchor=None, __shape=None, /, *, adjust_orientation=True, affine=True, axes=[1, 0], axis_names='WH', bytes_per_sample_hint=[0], device_memory_padding=16777216, device_memory_padding_jpeg2k=0, dtype=DALIDataType.UINT8, end=None, host_memory_padding=8388608, host_memory_padding_jpeg2k=0, hw_decoder_load=0.9, hybrid_huffman_threshold=1000000, jpeg_fancy_upsampling=False, normalized_anchor=True, normalized_shape=True, output_type=DALIImageType.RGB, preallocate_height_hint=0, preallocate_width_hint=0, preserve=False, rel_end=None, rel_shape=None, rel_start=None, seed=-1, shape=None, start=None, use_fast_idct=False, device=None, name=None)

Decodes images and extracts regions of interest.

Supported formats: JPEG, JPEG 2000, TIFF, PNG, BMP, PNM, PPM, PGM, PBM, WebP.

The output of the decoder is in HWC layout.

The implementation uses NVIDIA nvImageCodec to decode images. You need to install it separately. See https://developer.nvidia.com/nvimgcodec-downloads or simply do pip install nvidia-nvimgcodec-cu${CUDA_MAJOR_VERSION} where CUDA_MAJOR_VERSION is your CUDA major version (e.g. 12).

The slice can be specified by proving the start and end coordinates, or start coordinates and shape of the slice. Both coordinates and shapes can be provided in absolute or relative terms.

The slice arguments can be specified by the following named arguments:

  1. start: Slice start coordinates (absolute)

  2. rel_start: Slice start coordinates (relative)

  3. end: Slice end coordinates (absolute)

  4. rel_end: Slice end coordinates (relative)

  5. shape: Slice shape (absolute)

  6. rel_shape: Slice shape (relative)

The slice can be configured by providing start and end coordinates or start and shape. Relative and absolute arguments can be mixed (for example, rel_start can be used with shape) as long as start and shape or end are uniquely defined.

Alternatively, two extra positional inputs can be provided, specifying anchor and shape. When using positional inputs, two extra boolean arguments normalized_anchor/normalized_shape can be used to specify the nature of the arguments provided. Using positional inputs for anchor and shape is incompatible with the named arguments specified above.

The slice arguments should provide as many dimensions as specified by the axis_names or axes arguments.

By default, the nvidia.dali.fn.decoders.image_slice() operator uses normalized coordinates and “WH” order for the slice arguments.

When possible, the operator uses the ROI decoding, reducing the decoding time and memory consumption.

Note

GPU accelerated decoding is only available for a subset of the image formats (JPEG, and JPEG2000). For other formats, a CPU based decoder is used. For JPEG, a dedicated HW decoder will be used when available.

Note

WebP decoding currently only supports the simple file format (lossy and lossless compression). For details on the different WebP file formats, see https://developers.google.com/speed/webp/docs/riff_container

Supported backends

  • ‘cpu’

  • ‘mixed’

Parameters:

  • __data (TensorList) – Batch that contains the input data.

  • __anchor (1D TensorList of float or int, optional) –

    Input that contains normalized or absolute coordinates for the starting point of the slice (x0, x1, x2, …).

    Integer coordinates are interpreted as absolute coordinates, while float coordinates can be interpreted as absolute or relative coordinates, depending on the value of normalized_anchor.

  • __shape (1D TensorList of float or int, optional) –

    Input that contains normalized or absolute coordinates for the dimensions of the slice (s0, s1, s2, …).

    Integer coordinates are interpreted as absolute coordinates, while float coordinates can be interpreted as absolute or relative coordinates, depending on the value of normalized_shape.

Keyword Arguments:

  • adjust_orientation (bool, optional, default = True) – Use EXIF orientation metadata to rectify the images

  • affine (bool, optional, default = True) –

    Applies only to the mixed backend type.

    If set to True, each thread in the internal thread pool will be tied to a specific CPU core. Otherwise, the threads can be reassigned to any CPU core by the operating system.

  • axes (int or list of int or TensorList of int, optional, default = [1, 0]) –

    Order of dimensions used for the anchor and shape slice inputs as dimension indices.

    Negative values are interpreted as counting dimensions from the back. Valid range: [-ndim, ndim-1], where ndim is the number of dimensions in the input data.

  • axis_names (layout str, optional, default = ‘WH’) –

    Order of the dimensions used for the anchor and shape slice inputs, as described in layout.

    If a value is provided, axis_names will have a higher priority than axes.

  • bytes_per_sample_hint (int or list of int, optional, default = [0]) –

    Output size hint, in bytes per sample.

    If specified, the operator’s outputs residing in GPU or page-locked host memory will be preallocated to accommodate a batch of samples of this size.

  • device_memory_padding (int, optional, default = 16777216) –

    Applies only to the mixed backend type.

    The padding for nvJPEG’s device memory allocations, in bytes. This parameter helps to avoid reallocation in nvJPEG when a larger image is encountered, and the internal buffer needs to be reallocated to decode the image.

    If a value greater than 0 is provided, the operator preallocates one device buffer of the requested size per thread. If the value is correctly selected, no additional allocations will occur during the pipeline execution.

  • device_memory_padding_jpeg2k (int, optional, default = 0) –

    Applies only to the mixed backend type.

    The padding for nvJPEG2k’s device memory allocations, in bytes. This parameter helps to avoid reallocation in nvJPEG2k when a larger image is encountered, and the internal buffer needs to be reallocated to decode the image.

    If a value greater than 0 is provided, the operator preallocates the necessary number of buffers according to the hint provided. If the value is correctly selected, no additional allocations will occur during the pipeline execution.

  • dtype (nvidia.dali.types.DALIDataType, optional, default = DALIDataType.UINT8) –

    Output data type of the image.

    Values will be converted to the dynamic range of the requested type.

  • end (int or list of int or TensorList of int, optional) –

    End coordinates of the slice.

    Note: Providing named arguments start, end, shape, rel_start, rel_end, rel_shape is incompatible with providing positional inputs anchor and shape.

  • host_memory_padding (int, optional, default = 8388608) –

    Applies only to the mixed backend type.

    The padding for nvJPEG’s host memory allocations, in bytes. This parameter helps to prevent the reallocation in nvJPEG when a larger image is encountered, and the internal buffer needs to be reallocated to decode the image.

    If a value greater than 0 is provided, the operator preallocates two (because of double-buffering) host-pinned buffers of the requested size per thread. If selected correctly, no additional allocations will occur during the pipeline execution.

  • host_memory_padding_jpeg2k (int, optional, default = 0) –

    Applies only to the mixed backend type.

    The padding for nvJPEG2k’s host memory allocations, in bytes. This parameter helps to prevent the reallocation in nvJPEG2k when a larger image is encountered, and the internal buffer needs to be reallocated to decode the image.

    If a value greater than 0 is provided, the operator preallocates the necessary number of buffers according to the hint provided. If the value is correctly selected, no additional allocations will occur during the pipeline execution.

  • hw_decoder_load (float, optional, default = 0.9) –

    The percentage of the image data to be processed by the HW JPEG decoder.

    Applies only to the mixed backend type in NVIDIA Ampere GPU architecture.

    Determines the percentage of the workload that will be offloaded to the hardware decoder, if available. The optimal workload depends on the number of threads that are provided to the DALI pipeline and should be found empirically. More details can be found at https://developer.nvidia.com/blog/loading-data-fast-with-dali-and-new-jpeg-decoder-in-a100

  • hybrid_huffman_threshold (int, optional, default = 1000000) –

    Applies only to the mixed backend type.

    Images with a total number of pixels (height * width) that is higher than this threshold will use the nvJPEG hybrid Huffman decoder. Images that have fewer pixels will use the nvJPEG host-side Huffman decoder.

    Note

    Hybrid Huffman decoder still largely uses the CPU.

  • jpeg_fancy_upsampling (bool, optional, default = False) –

    Make the mixed backend use the same chroma upsampling approach as the cpu one.

    The option corresponds to the JPEG fancy upsampling available in libjpegturbo or ImageMagick.

  • memory_stats –

normalized_anchorbool, optional, default = True

Determines whether the anchor positional input should be interpreted as normalized (range [0.0, 1.0]) or as absolute coordinates.

Note

This argument is only relevant when anchor data type is float. For integer types, the coordinates are always absolute.

normalized_shapebool, optional, default = True

Determines whether the shape positional input should be interpreted as normalized (range [0.0, 1.0]) or as absolute coordinates.

Note

This argument is only relevant when anchor data type is float. For integer types, the coordinates are always absolute.

output_typenvidia.dali.types.DALIImageType, optional, default = DALIImageType.RGB

The color space of the output image.

Note: When decoding to YCbCr, the image will be decoded to RGB and then converted to YCbCr, following the YCbCr definition from ITU-R BT.601.

preallocate_height_hintint, optional, default = 0

Image width hint.

Applies only to the mixed backend type in NVIDIA Ampere GPU architecture.

The hint is used to preallocate memory for the HW JPEG decoder.

preallocate_width_hintint, optional, default = 0

Image width hint.

Applies only to the mixed backend type in NVIDIA Ampere GPU architecture.

The hint is used to preallocate memory for the HW JPEG decoder.

preservebool, optional, default = False

Prevents the operator from being removed from the graph even if its outputs are not used.

rel_endfloat or list of float or TensorList of float, optional

End relative coordinates of the slice (range [0.0 - 1.0].

Note: Providing named arguments start, end, shape, rel_start, rel_end, rel_shape is incompatible with providing positional inputs anchor and shape.

rel_shapefloat or list of float or TensorList of float, optional

Relative shape of the slice (range [0.0 - 1.0]).

Providing named arguments start, end, shape, rel_start, rel_end, rel_shape is incompatible with providing positional inputs anchor and shape.

rel_startfloat or list of float or TensorList of float, optional

Start relative coordinates of the slice (range [0.0 - 1.0]).

Note: Providing named arguments start, end, shape, rel_start, rel_end, rel_shape is incompatible with providing positional inputs anchor and shape.

seedint, optional, default = -1

Random seed.

If not provided, it will be populated based on the global seed of the pipeline.

shapeint or list of int or TensorList of int, optional

Shape of the slice.

Providing named arguments start, end, shape, rel_start, rel_end, rel_shape is incompatible with providing positional inputs anchor and shape.

split_stages : bool, optional, default = False

Warning

The argument split_stages is now deprecated and its usage is discouraged.

startint or list of int or TensorList of int, optional

Start coordinates of the slice.

Note: Providing named arguments start/end or start/shape is incompatible with providing positional inputs anchor and shape.

use_chunk_allocator : bool, optional, default = False

Warning

The argument use_chunk_allocator is now deprecated and its usage is discouraged.

use_fast_idctbool, optional, default = False

Enables fast IDCT in the libjpeg-turbo based CPU decoder, used when device is set to “cpu” or when the it is set to “mixed” but the particular image can not be handled by the GPU implementation.

According to the libjpeg-turbo documentation, decompression performance is improved by up to 14% with little reduction in quality.