nvidia.dali.fn.experimental.debayer#

nvidia.dali.fn.experimental.debayer(__input, /, *, blue_position, algorithm='bilinear_npp', bytes_per_sample_hint=[0], preserve=False, seed=-1, device=None, name=None)#

Performs image demosaicing/debayering.

Converts single-channel image to RGB using specified color filter array.

The supported input types are uint8_t and uint16_t. The input images must be 2D tensors (HW) or 3D tensors (HWC) where the number of channels is 1. The operator supports sequence of images/video-like inputs (layout FHW).

For example, the following snippet presents debayering of batch of image sequences:

def bayered_sequence(sample_info):
  # some actual source of video inputs with corresponding pattern
  # as opencv-style string
  video, bayer_pattern = get_sequence(sample_info)
  if bayer_pattern == "bggr":
      blue_position = [0, 0]
  elif bayer_pattern == "gbrg":
      blue_position = [0, 1]
  elif bayer_pattern == "grbg":
      blue_position = [1, 0]
  else:
      assert bayer_pattern == "rggb"
      blue_position = [1, 1]
  return video, np.array(blue_position, dtype=np.int32)

@pipeline_def
def debayer_pipeline():
  bayered_sequences, blue_positions = fn.external_source(
    source=bayered_sequence, batch=False, num_outputs=2,
    layout=["FHW", None])  # note the "FHW" layout, for plain images it would be "HW"
  debayered_sequences = fn.experimental.debayer(
    bayered_sequences.gpu(), blue_position=blue_positions)
  return debayered_sequences

This operator allows sequence inputs.

Supported backends
  • ‘gpu’

Parameters:

__input (TensorList ('HW', 'HWC', 'FHW', 'FHWC')) – Input to the operator.

Keyword Arguments:
  • blue_position (int or list of int or TensorList of int) –

    The layout of color filter array/bayer tile.

    A position of the blue value in the 2x2 bayer tile. The supported values correspond to the following OpenCV bayer layouts:

    • (0, 0) - BG/BGGR

    • (0, 1) - GB/GBRG

    • (1, 0) - GR/GRBG

    • (1, 1) - RG/RGGB

    The argument follows OpenCV’s convention of referring to a 2x2 tile that starts in the second row and column of the sensors’ matrix.

    For example, the (0, 0)/BG/BGGR corresponds to the following matrix of sensors:

    R

    G

    R

    G

    R

    G

    B

    G

    B

    G

    R

    G

    R

    G

    R

    G

    B

    G

    B

    G

    Supports per-frame inputs.

  • algorithm (str, optional, default = ‘bilinear_npp’) –

    The algorithm to be used when inferring missing colours for any given pixel. Currently only bilinear_npp is supported.

    • The bilinear_npp algorithm uses bilinear interpolation to infer red and blue values. For green values a bilinear interpolation with chroma correlation is used as explained in NPP documentation.

  • 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.

  • preserve (bool, optional, default = False) – Prevents the operator from being removed from the graph even if its outputs are not used.

  • seed (int, optional, default = -1) –

    Random seed.

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