def get_crop_hints(v_client, filepath):
"""Given a JPG location to open, retrieves vertices to crop to.
Args:
v_client (google.cloud.vision_v1.ImageAnnotatorClient)
filepath (str): 'path/to/assets/name.jpg'
Returns:
list-esque google.protobuf.internal.containers.RepeatedCompositeFieldContainer
of four google.cloud.vision_v1.types.Vertex objects. Individual points
can be accessed like:
verts[0].x, verts[0].y,
verts[1].x, verts[1].y,
verts[2].x, verts[2].y,
verts[3].x, verts[3].y
"""
# https://cloud.google.com/vision/docs/crop-hints
image = vision_img_from_path(v_client, filepath)
response = v_client.crop_hints(image=image)
logger.debug(f"API response for crop_hints: {response}")
if not response.crop_hints_annotation:
logger.error(f"No crop hints annotation for image.")
raise exceptions.GoogleAPIError(f"No object annotations for image. Vision API response: {response}")
hints = response.crop_hints_annotation.crop_hints
vertices = hints[0].bounding_poly.vertices
logger.debug(f"Returning crop hints bounding poly vertices: {vertices}")
return vertices
def get_safety_annotations(v_client, image):
# TODO: Change signature to match get_object_annotations and get_label_annotations.
"""
Args:
v_client (google.cloud.vision_v1.ImageAnnotatorClient)
image (google.cloud.vision_v1.types.Image
Returns:
dict of likelihoods that image contains an unsafe category:
{'adult': 'VERY_UNLIKELY', 'medical': 'UNLIKELY', 'spoofed': 'POSSIBLE',
'violence': 'LIKELY', 'racy': 'VERY_LIKELY'}
"""
# https://cloud.google.com/vision/docs/detecting-safe-search
response = v_client.safe_search_detection(image=image)
logger.debug(f"API response for safe_search_detection: {response}")
if not response.safe_search_annotation:
logger.error(f"No safety annotations for image.")
raise exceptions.GoogleAPIError(f"No safety annotations for image. Vision API response: {response}")
likelihood_name = ('UNKNOWN', 'VERY_UNLIKELY', 'UNLIKELY', 'POSSIBLE',
'LIKELY', 'VERY_LIKELY')
safety_annotations = {"adult": likelihood_name[response.safe_search_annotation.adult],
"medical": likelihood_name[response.safe_search_annotation.medical],
"spoofed": likelihood_name[response.safe_search_annotation.spoof],
"violence": likelihood_name[response.safe_search_annotation.violence],
"racy": likelihood_name[response.safe_search_annotation.racy]}
logger.debug(f"Returning safety_annotations: {safety_annotations}")
return safety_annotations
def get_crop_hints(v_client, filepath):
# https://cloud.google.com/vision/docs/crop-hints
image = vision_img_from_path(v_client, filepath)
response = v_client.crop_hints(image=image)
logger.debug(f"API response for crop_hints: {response}")
if not response.crop_hints_annotation:
logger.error(f"No crop hints annotation for image.")
raise exceptions.GoogleAPIError(
f"No object annotations for image. Vision API response: {response}"
)
hints = response.crop_hints_annotation.crop_hints
vertices = hints[0].bounding_poly.vertices
logger.debug(f"Returning crop hints bounding poly vertices: {vertices}")
return vertices
def get_object_annotations(v_client, filepath):
"""Given a JPG location to open, return information about objects detected
in the image.
Args:
v_client (google.cloud.vision_v1.ImageAnnotatorClient)
filepath (str): 'path/to/assets/name.jpg'
Returns:
list of object annotation dictionaries with object name and box
dimensions:
[{'name': 'bird', 'crop_box': [392, 353, 542, 470], 'draw_box': [392,
353, 542, 353, 542, 470, 392, 470]},
{'name': 'animal', 'crop_box': [392, 353, 542, 470], 'draw_box': [392,
353, 542, 353, 542, 470, 392, 470]}]
"""
# https://cloud.google.com/vision/docs/detecting-objects
image = vision_img_from_path(v_client, filepath)
response = v_client.object_localization(image=image)
logger.debug(f"Response for {filepath} object_localization request: {response}")
if not response.localized_object_annotations:
logger.error(f"No object annotations for {filepath}.")
raise exceptions.GoogleAPIError(f"No object annotations for {filepath}. Vision API response: {response}")
width, height = Image.open(filepath).size
object_annotations = list()
for o in response.localized_object_annotations:
oa = dict()
verts = o.bounding_poly.normalized_vertices
oa["name"] = o.name.lower()
# https://pillow.readthedocs.io/en/5.3.x/reference/Image.html#PIL.Image.Image.crop
oa["crop_box"] = [round(verts[0].x * width), # left
round(verts[0].y * height), # upper
round(verts[2].x * width), # right
round(verts[2].y * height)] # lower
# https://pillow.readthedocs.io/en/5.3.x/reference/ImageDraw.html#PIL.ImageDraw.PIL.ImageDraw.ImageDraw.polygon
oa["draw_box"] = [round(verts[0].x * width), round(verts[0].y * height),
round(verts[1].x * width), round(verts[1].y * height),
round(verts[2].x * width), round(verts[2].y * height),
round(verts[3].x * width), round(verts[3].y * height)]
object_annotations.append(oa)
logger.debug(f"Returning object annotations: {object_annotations}")
return object_annotations
def get_label_annotations(v_client, filepath):
"""Given a JPG location to open, retrieves image's label annotations.
Args:
v_client (google.cloud.vision_v1.ImageAnnotatorClient)
filepath (str): 'path/to/assets/name.jpg'
Returns:
list of str labels that describe image contents: ['Bird', 'Soil', 'Lark']
"""
# https://cloud.google.com/vision/docs/reference/rest/v1/images/annotate#EntityAnnotation
image = vision_img_from_path(v_client, filepath)
response = v_client.label_detection(image=image)
logger.debug(f"API response for label_detection: {response}")
if not response.label_annotations:
logger.error(f"No label annotations for image.")
raise exceptions.GoogleAPIError(f"No label annotations for image. Vision API response: {response}")
labels = list(l.description for l in response.label_annotations)
logger.debug(f"Returning label annotations: {labels}")
return labels
def get_label_annotations(v_client, filepath):
# TODO: Improve docstring
"""Retrieve image's label annotations from Cloud Vision API.
Arg: google.cloud.vision_v1.types.Image
Returns:
labels (list): List of EntityAnnotation objects. E.g., [mid: "/m/02mhj"
description: "ecosystem" score: 0.9368894100189209 topicality: 0.9368894100189209, ...]
See https://cloud.google.com/vision/docs/reference/rest/v1/images/annotate#EntityAnnotation
"""
image = vision_img_from_path(v_client, filepath)
response = v_client.label_detection(image=image)
logger.debug(f"API response for label_detection: {response}")
if not response.label_annotations:
logger.error(f"No label annotations for image.")
raise exceptions.GoogleAPIError(
f"No label annotations for image. Vision API response: {response}")
labels = list(l.description for l in response.label_annotations)
logger.debug(f"Returning label annotations: {labels}")
return labels
def get_object_annotations(v_client, filepath):
# TODO: Docstring
# https://cloud.google.com/vision/docs/detecting-objects
image = vision_img_from_path(v_client, filepath)
response = v_client.object_localization(image=image)
logger.debug(
f"Response for {filepath} object_localization request: {response}")
if not response.localized_object_annotations:
logger.error(f"No object annotations for {filepath}.")
raise exceptions.GoogleAPIError(
f"No object annotations for {filepath}. Vision API response: {response}"
)
width, height = Image.open(filepath).size
object_annotations = list()
for o in response.localized_object_annotations:
oa = dict()
verts = o.bounding_poly.normalized_vertices
oa["name"] = o.name.lower()
# https://pillow.readthedocs.io/en/5.3.x/reference/Image.html#PIL.Image.Image.crop
oa["crop_box"] = [
round(verts[0].x * width), # left
round(verts[0].y * height), # upper
round(verts[2].x * width), # right
round(verts[2].y * height)
] # lower
# https://pillow.readthedocs.io/en/5.3.x/reference/ImageDraw.html#PIL.ImageDraw.PIL.ImageDraw.ImageDraw.polygon
oa["draw_box"] = [
round(verts[0].x * width),
round(verts[0].y * height),
round(verts[1].x * width),
round(verts[1].y * height),
round(verts[2].x * width),
round(verts[2].y * height),
round(verts[3].x * width),
round(verts[3].y * height)
]
object_annotations.append(oa)
logger.debug(f"Returning object annotations: {object_annotations}")
return object_annotations
def get_safety_annotations(v_client, image):
# TODO: Docstring
""" arg google.cloud.vision_v1.types.Image"""
# https://cloud.google.com/vision/docs/detecting-safe-search
response = v_client.safe_search_detection(image=image)
logger.debug(f"API response for safe_search_detection: {response}")
if not response.safe_search_annotation:
logger.error(f"No safety annotations for image.")
raise exceptions.GoogleAPIError(
f"No safety annotations for image. Vision API response: {response}"
)
likelihood_name = ('UNKNOWN', 'VERY_UNLIKELY', 'UNLIKELY', 'POSSIBLE',
'LIKELY', 'VERY_LIKELY')
safety_annotations = {
"adult": likelihood_name[response.safe_search_annotation.adult],
"medical": likelihood_name[response.safe_search_annotation.medical],
"spoofed": likelihood_name[response.safe_search_annotation.spoof],
"violence": likelihood_name[response.safe_search_annotation.violence],
"racy": likelihood_name[response.safe_search_annotation.racy]
}
logger.debug(f"Returning safety_annotations: {safety_annotations}")
return safety_annotations