def test_show_embedded_jpeg(tmp_path):
data = np.random.random((100, 100))
rescaled = (255.0 / data.max() * (data - data.min())).astype(np.uint8)
im = PILImage.fromarray(rescaled)
uri = tmp_path / "test.jpg"
im.save(uri)
result = Image(uri)._repr_jpeg_()
with open(uri, "rb") as fh:
expected = b2a_base64(fh.read()).decode("ascii")
assert result == expected
fh.seek(0)
embedded_image = Image(fh)
assert result == embedded_image._repr_jpeg_()
def test_show_remote_ref():
from IPython.display import Image as IPyImage
uri = "https://octodex.github.com/images/original.png"
img = Image(uri)
# TODO check the actual content
assert img._repr_html_() == img.display()._repr_html_()
assert img.display()._repr_html_() == IPyImage(uri)._repr_html_()
def test_embeded_image_from_bytesio():
data = np.random.random((100, 100))
rescaled = (255.0 / data.max() * (data - data.min())).astype(np.uint8)
im = PILImage.fromarray(rescaled)
buf = BytesIO()
im.save(buf, format="PNG")
buf.seek(0)
image = Image(buf)
assert np.array_equal(image.to_numpy(), rescaled)
def test_image_use_https_uri():
img = Image(WIKIPEDIA)
fobj = BytesIO(
requests.get(WIKIPEDIA, headers={
"User-Agent": "curl/7.72.0"
}).content)
pic = PIL.Image.open(fobj)
assert np.array_equal(img.to_numpy(), np.array(pic))
def test_crop_in_batch():
uri = "http://farm2.staticflickr.com/1129/4726871278_4dd241a03a_z.jpg"
img = Image(uri)
data = img.to_numpy()
patches = img.crop(
[Box2d(10, 10, 30, 30),
Box2d(15, 15, 35, 35),
Box2d(20, 20, 40, 40)])
assert len(patches) == 3
assert np.array_equal(patches[0].to_numpy(), data[10:30, 10:30, :])
assert np.array_equal(patches[1].to_numpy(), data[15:35, 15:35, :])
assert np.array_equal(patches[2].to_numpy(), data[20:40, 20:40, :])
def image_copy(img: Image, uri: str) -> Image:
"""Copy the image to a new destination, specified by the URI.
Parameters
----------
img : Image
An image object
uri : str
The base directory to copy the image to.
Return
------
Image
Return a new image pointed to the new URI
"""
logger.info("Copying image src=%s dest=%s", img.uri, uri)
return Image(_copy(img.uri, uri))
def to_image(image_data: Union[bytes, bytearray, str, Path]) -> Image:
"""Build an :py:class:`Image` from
bytes, file-like object, str, or :py:class:`~pathlib.Path`.
Parameters
----------
image_data : bytes, bytearray, str, Path
The resource identifier or bytes of the source image.
Return
------
img: Image
An Image from the given embedded data or URI
Example
-------
>>> df = spark.read.format("image").load("<path-to-data>")
>>>
>>> df.withColumn("new_image", to_image("image.data"))
"""
return Image(image_data)
def deserialize(self, datum) -> "Image":
from rikai.types.vision import Image # pylint: disable=import-outside-toplevel
return Image(datum[0])
def image(uri: str) -> Image:
"""Build an :py:class:`Image` from a URI."""
return Image(uri)
def deserialize(self, datum) -> "Image":
from rikai.types.vision import Image
return Image(datum[0] or datum[1])
def test_crop_real_image():
uri = "http://farm2.staticflickr.com/1129/4726871278_4dd241a03a_z.jpg"
img = Image(uri)
data = img.to_numpy()
patch = img.crop(Box2d(10, 10, 30, 30))
assert np.array_equal(patch.to_numpy(), data[10:30, 10:30, :])
def uri_to_pil(uri):
# TODO: We can remove this after UDT is supported in Spark
return Image(uri).to_pil()
def test_image_use_https_uri():
img = Image(WIKIPEDIA)
fobj = BytesIO(requests.get(WIKIPEDIA).content)
pic = PIL.Image.open(fobj)
assert np.array_equal(img.to_numpy(), np.array(pic))