mediapipe_model_maker.face_stylizer.dataset.image_module.Image

A container for storing an image or a video frame, in one of several formats.

mediapipe_model_maker.face_stylizer.dataset.image_module.Image()

Formats supported by Image are listed in the ImageFormat enum. Pixels are encoded row-major in an interleaved fashion. Image supports uint8, uint16, and float as its data types.

Image can be created by copying the data from a numpy ndarray that stores the pixel data continuously. An Image may realign the input data on its default alignment boundary during creation. The data in an Image will become immutable after creation.

Creation examples:

import cv2
cv_mat = cv2.imread(input_file)
rgb_frame = mp.Image(image_format=mp.ImageFormat.SRGB, data=cv_mat)
gray_frame = mp.Image(
    image_format=mp.ImageFormat.GRAY8,
    data=cv2.cvtColor(cv_mat, cv2.COLOR_RGB2GRAY))

from PIL import Image
pil_img = Image.new('RGB', (60, 30), color = 'red')
image = mp.Image(
    image_format=mp.ImageFormat.SRGB, data=np.asarray(pil_img))

The pixel data in an Image can be retrieved as a numpy ndarray by calling Image.numpy_view(). The returned numpy ndarray is a reference to the internal data and itself is unwritable. If the callers want to modify the numpy ndarray, it's required to obtain a copy of it.

Pixel data retrieval examples:

for channel in range(num_channel):
  for col in range(width):
    for row in range(height):
      print(image[row, col, channel])

output_ndarray = image.numpy_view()
print(output_ndarray[0, 0, 0])
copied_ndarray = np.copy(output_ndarray)
copied_ndarray[0,0,0] = 0

Methods

create_from_file

create_from_file()

Creates Image object from the image file.

is_aligned

is_aligned()

Return True if each row of the data is aligned to alignment boundary, which must be 1 or a power of 2.

image.is_aligned(16)

is_contiguous

is_contiguous()

Return True if the pixel data is stored contiguously (without any alignment padding areas).

is_empty

is_empty()

Return True if the pixel data is unallocated.

numpy_view

numpy_view()

Return the image pixel data as an unwritable numpy ndarray.

Realign the pixel data to be stored contiguously and return a reference to the unwritable numpy ndarray. If the callers want to modify the numpy array data, it's required to obtain a copy of the ndarray.

output_ndarray = image.numpy_view()
copied_ndarray = np.copy(output_ndarray)
copied_ndarray[0,0,0] = 0

uses_gpu

uses_gpu()

Return True if data is currently on the GPU.

__getitem__

__getitem__()

Use the indexer operators to access pixel data.

for channel in range(num_channel):
  for col in range(width):
    for row in range(height):
      print(image[row, col, channel])