def test_should_tag_single_token_within_partial_prediction_at_smaller_scale(self):
token_1 = SimpleToken(TOKEN_TEXT_1)
structured_document = SimpleStructuredDocument(lines=[SimpleLine([token_1])])
structured_document.set_bounding_box(
structured_document.get_pages()[0],
BoundingBox(0, 0, DEFAULT_WIDTH * 100, DEFAULT_HEIGHT * 100)
)
structured_document.set_bounding_box(
token_1,
BoundingBox(0, 0, DEFAULT_WIDTH * 10, DEFAULT_HEIGHT * 10)
)
annotated_image = filled_image(
BG_COLOR, {TAG_1: COLOR_1},
width=DEFAULT_WIDTH * 10,
height=DEFAULT_HEIGHT * 10
)
fill_rect(
annotated_image,
BoundingBox(0, 0, DEFAULT_WIDTH, DEFAULT_HEIGHT),
COLOR_1
)
annotate_structured_document_using_predicted_images(
structured_document,
[annotated_image]
)
assert structured_document.get_tag(token_1, scope=CV_TAG_SCOPE) == TAG_1
def test_should_calculate_bounding_box_of_page_without_xy(self):
page = E.PAGE({
'width': '100',
'height': '101'
})
doc = LxmlStructuredDocument(E.DOCUMENT(page))
assert doc.get_bounding_box(page) == BoundingBox(0, 0, 100, 101)
def test_should_be_able_to_set_bounding_box(self):
bounding_box = BoundingBox(11, 12, 101, 102)
text = SVG_TEXT('a', {'x': '10', 'y': '11', 'font-size': '100'})
doc = SvgStructuredDocument(E.svg(SVG_TEXT_LINE(text)))
doc.set_bounding_box(text, bounding_box)
assert text.attrib[SVGE_BOUNDING_BOX] == format_bounding_box(
bounding_box)
def get_node_bounding_box(t):
return BoundingBox(
float(t.attrib.get('x', 0)),
float(t.attrib.get('y', 0)),
float(t.attrib['width']),
float(t.attrib['height'])
)
def test_should_use_viewbox_if_available(self):
bounding_box = BoundingBox(11, 12, 101, 102)
page = E.svg({
SVG_VIEWBOX_ATTRIB: format_bounding_box(bounding_box)
})
doc = SvgStructuredDocument(page)
assert doc.get_bounding_box(page) == bounding_box
def test_should_create_text_node_with_common_attributes(self):
lxml_root = E.DOCUMENT(
E.PAGE(
E.TEXT(
E.TOKEN(
SOME_TEXT,
COMMON_LXML_TOKEN_ATTRIBS
)
)
)
)
svg_pages = list(iter_svg_pages_for_lxml(lxml_root))
assert len(svg_pages) == 1
first_page = svg_pages[0]
svg_text = first_page.find('.//' + SVG_TEXT)
assert svg_text is not None
assert svg_text.text == SOME_TEXT
assert float(svg_text.attrib[SVG.X]) == float(SOME_X)
assert float(svg_text.attrib[SVG.Y]) == float(SOME_Y)
assert float(svg_text.attrib[SVG.FONT_SIZE]) == float(SOME_FONT_SIZE)
assert svg_text.attrib[SVG.FONT_FAMILY] == SOME_FONT_FAMILY
assert svg_text.attrib[SVG.FILL] == SOME_FONT_COLOR
assert parse_bounding_box(svg_text.attrib.get(SVG.BOUNDING_BOX)) == BoundingBox(
float(COMMON_LXML_TOKEN_ATTRIBS[LXML.X]),
float(COMMON_LXML_TOKEN_ATTRIBS[LXML.Y]),
float(COMMON_LXML_TOKEN_ATTRIBS[LXML.WIDTH]),
float(COMMON_LXML_TOKEN_ATTRIBS[LXML.HEIGHT])
)
def test_should_accept_float_image_size(self):
blocks = [AnnotationBlock(TAG1, BoundingBox(0, 0, 1, 1))]
image = annotated_blocks_to_image(blocks,
color_map={TAG1: (0, 255, 0)},
width=3.1,
height=3.9)
assert image.size == (4, 4)
def test_should_convert_rect_color_name(self):
blocks = [AnnotationBlock(TAG1, BoundingBox(0, 0, 1, 1))]
image = annotated_blocks_to_image(blocks,
color_map={TAG1: 'green'},
width=3,
height=3)
assert image.getpixel((0, 0)) == (0, 128, 0, 255)
def test_should_ignore_unmapped_tag(self):
blocks = [AnnotationBlock(TAG1, BoundingBox(0, 0, 1, 1))]
image = annotated_blocks_to_image(blocks,
color_map={},
width=3,
height=3)
assert image.getpixel((0, 0)) == (255, 255, 255, 0)
def test_should_calculate_default_bounding_box(self):
text = SVG_TEXT('a', {
'x': '10',
'y': '11',
'font-size': '100'
})
doc = SvgStructuredDocument(E.svg(SVG_TEXT_LINE(text)))
assert doc.get_bounding_box(text) == BoundingBox(10, 11, 100 * 0.8, 100)
def test_should_create_rect_for_single_annotated_block(self):
blocks = [AnnotationBlock(TAG1, BoundingBox(0, 0, 1, 1))]
image = annotated_blocks_to_image(blocks,
color_map={TAG1: (0, 255, 0)},
width=3,
height=3)
assert image.getpixel((0, 0)) == (0, 255, 0, 255)
def test_should_calculate_default_bounding_box(self):
token = E.TOKEN({
'x': '10',
'y': '11',
'width': '100',
'height': '101'
})
doc = LxmlStructuredDocument(E.DOCUMENT(E.PAGE(E.TEXT(token))))
assert doc.get_bounding_box(token) == BoundingBox(10, 11, 100, 101)
def test_should_use_bounding_box_if_available(self):
bounding_box = BoundingBox(11, 12, 101, 102)
text = SVG_TEXT('a', {
'x': '10',
'y': '11',
'font-size': '100',
SVGE_BOUNDING_BOX: format_bounding_box(bounding_box)
})
doc = SvgStructuredDocument(E.svg(SVG_TEXT_LINE(text)))
assert doc.get_bounding_box(text) == bounding_box
def test_should_be_able_to_set_bounding_box(self):
bounding_box = BoundingBox(10, 11, 100, 101)
token = E.TOKEN({
'x': '20',
'y': '21',
'width': '200',
'height': '201'
})
doc = LxmlStructuredDocument(E.DOCUMENT(E.PAGE(E.TEXT(token))))
doc.set_bounding_box(token, bounding_box)
assert doc.get_bounding_box(token) == bounding_box
def test_should_estimate_width_based_on_number_of_characters(self):
s = 'abc'
text = SVG_TEXT(s, {
'x': '10',
'y': '11',
'font-size': '100'
})
doc = SvgStructuredDocument(E.svg(SVG_TEXT_LINE(text)))
assert doc.get_bounding_box(text) == BoundingBox(
10, 11, 100 * 0.8 * len(s), 100
)
def get_node_bounding_box(t):
attrib = t.attrib
if SVGE_BOUNDING_BOX in attrib:
return parse_bounding_box(attrib[SVGE_BOUNDING_BOX])
if SVG_VIEWBOX_ATTRIB in attrib:
return parse_bounding_box(attrib[SVG_VIEWBOX_ATTRIB])
if not ('font-size' in attrib and 'x' in attrib and 'y' in attrib):
return None
font_size = float(attrib['font-size'])
width = font_size * 0.8 * max(1, len(t.text))
return BoundingBox(float(attrib['x']), float(attrib['y']), width,
font_size)
def test_should_merge_right_nearby_block_with_same_tag_using_fractions(
self):
block1 = AnnotationBlock(TAG1, BoundingBox(10.5, 10.5, 10.9, 26.3))
block2 = AnnotationBlock(
TAG1,
block1.bounding_box.move_by(
block1.bounding_box.width + DEFAULT_NEARBY_TOLERANCE, 0))
merged_blocks = merge_blocks([block1, block2],
nearby_tolerance=DEFAULT_NEARBY_TOLERANCE)
assert [b.tag for b in merged_blocks] == [TAG1]
assert merged_blocks[0].bounding_box == block1.bounding_box.include(
block2.bounding_box)
from sciencebeam_gym.structured_document.svg import (SVG_NS)
from sciencebeam_gym.preprocess.blockify_annotations import (
annotation_document_page_to_annotation_blocks,
annotation_document_page_to_merged_blocks, annotated_blocks_to_svg,
annotated_blocks_to_image, merge_blocks, AnnotationBlock)
TAG1 = 'tag1'
TAG2 = 'tag2'
DEFAULT_SVGX_WIDTH = 10
DEFAULT_FONT_SIZE = 10
DEFAULT_COLOR = 'red'
DEFAULT_BOUNDING_BOX = BoundingBox(0, 0, 16, 10)
DEFAULT_NEARBY_TOLERANCE = 10
def setup_module():
logging.basicConfig(level=logging.DEBUG)
class TestAnnotatedBlocksToSvg(object):
def test_should_create_rect_for_single_annotated_block(self):
blocks = [AnnotationBlock(TAG1, DEFAULT_BOUNDING_BOX)]
result_svg = annotated_blocks_to_svg(blocks,
color_map={TAG1: DEFAULT_COLOR},
width=100,
def test_should_not_equal_none(self):
assert not BoundingBox(11, 12, 101, 102).__eq__(None)
def test_should_include_another_bounding_box_to_the_top_left(self):
assert (BoundingBox(100, 100, 200, 200).include(
BoundingBox(10, 20, 50,
100)) == BoundingBox(10, 20, 100 + 200 - 10,
100 + 200 - 20))
def test_should_indicate_empty_with_zero_height(self):
assert BoundingBox(0, 0, 100, 0).empty()
def parse_bounding_box(bounding_box_str):
if not bounding_box_str:
return None
x, y, width, height = bounding_box_str.split()
return BoundingBox(float(x), float(y), float(width), float(height))
SimplePage,
SimpleLine,
SimpleToken
)
from sciencebeam_gym.models.text.feature_extractor import (
structured_document_to_token_props,
token_props_list_to_features,
token_props_list_to_labels,
remove_labels_from_token_props_list,
merge_with_cv_structured_document,
NONE_TAG,
CV_TAG_SCOPE
)
PAGE_BOUNDING_BOX = BoundingBox(0, 0, 100, 200)
TOKEN_BOUNDING_BOX = BoundingBox(10, 10, 10, 20)
TEXT_1 = 'Text 1'
TEXT_2 = 'Text 2'
TEXT_3 = 'Text 3'
TAG_1 = 'tag1'
TAG_2 = 'tag2'
TAG_3 = 'tag3'
SCOPE_1 = 'scope1'
class TestStructuredDocumentToTokenProps(object):
def test_should_return_empty_token_list_if_document_has_no_pages(self):
def scale_bounding_box(bounding_box, rx, ry):
return BoundingBox(bounding_box.x * rx, bounding_box.y * ry,
bounding_box.width * rx, bounding_box.height * ry)
annotate_structured_document_using_predicted_images,
parse_args,
CV_TAG_SCOPE
)
TAG_1 = 'tag1'
TOKEN_TEXT_1 = 'a token value'
COLOR_1 = (1, 1, 1)
BG_COLOR = (255, 255, 255)
DEFAULT_HEIGHT = 3
DEFAULT_WIDTH = 3
DEFAULT_BOUNDING_BOX = BoundingBox(0, 0, DEFAULT_WIDTH, DEFAULT_HEIGHT)
def filled_image(color, color_map, width=DEFAULT_WIDTH, height=DEFAULT_HEIGHT):
return AnnotatedImage(
np.full((height, width, 3), color),
color_map
)
def fill_rect(annoted_image, bounding_box, color):
for y in range(bounding_box.y, bounding_box.y + bounding_box.height):
for x in range(bounding_box.x, bounding_box.x + bounding_box.width):
annoted_image.data[y, x] = color
def test_should_return_zero_tag_probality_if_bounding_box_is_empty(self):
annotated_image = filled_image(BG_COLOR, {TAG_1: COLOR_1})
assert annotated_image.get_tag_probabilities_within(
BoundingBox(0, 0, 0, 0)
).get(TAG_1) == 0.0
def iter_svg_pages_for_lxml(lxml_root, add_background=True):
previous_block = None
previous_svg_block = None
for page in lxml_root.xpath('//DOCUMENT/PAGE'):
svg_root = etree.Element(SVG_DOC, nsmap=SVG_NSMAP)
page_width = page.attrib.get('width')
page_height = page.attrib.get('height')
if page_width and page_height:
svg_root.attrib['viewBox'] = '0 0 %s %s' % (page_width,
page_height)
if add_background:
svg_root.append(
etree.Element(SVG_RECT,
attrib={
'width': '100%',
'height': '100%',
'fill': 'white',
'class': 'background'
}))
for text in page.xpath('.//TEXT'):
svg_g = etree.Element(SVG_G,
nsmap=SVG_NSMAP,
attrib={'class': SvgStyleClasses.LINE})
for token in text.xpath('./TOKEN'):
x = float(token.attrib.get('x'))
y = float(token.attrib.get('y'))
height = float(token.attrib.get('height'))
width = float(token.attrib.get('width'))
base = float(token.attrib.get('base', y))
y_center = y + height / 2.0
attrib = {
'x':
str(x),
'y':
str(base),
'font-size':
token.attrib.get('font-size'),
'font-family':
token.attrib.get('font-name'),
'fill':
token.attrib.get('font-color'),
SVGE_BOUNDING_BOX:
svg_format_bounding_box(BoundingBox(x, y, width, height))
}
angle = float(token.attrib.get('angle', '0'))
if token.attrib.get('rotation') == '1' and angle == 90.0:
attrib['x'] = '0'
attrib['y'] = '0'
attrib[
'transform'] = 'translate({x} {y}) rotate({angle})'.format(
x=str(x), y=str(y_center), angle=str(-angle))
svg_g.append(
ElementWithText(SVG_TEXT, token.text, attrib=attrib))
text_parent = text.getparent()
if text_parent.tag == 'BLOCK':
if text_parent != previous_block:
previous_svg_block = etree.Element(
SVG_G,
nsmap=SVG_NSMAP,
attrib={'class': SvgStyleClasses.BLOCK})
svg_root.append(previous_svg_block)
previous_block = text_parent
previous_svg_block.append(svg_g)
else:
previous_block = None
previous_svg_block = None
svg_root.append(svg_g)
yield svg_root
def test_should_not_equal_bounding_boxes_with_different_height(self):
assert BoundingBox(11, 12, 101, 102) != BoundingBox(11, 12, 101, 999)
from sciencebeam_gym.models.text.feature_extractor import (
structured_document_to_token_props, token_props_list_to_features, NONE_TAG)
from sciencebeam_gym.utils.bounding_box import (BoundingBox)
from sciencebeam_gym.models.text.crf.annotate_using_predictions import (
annotate_structured_document_using_predictions,
predict_and_annotate_structured_document, CRF_TAG_SCOPE)
TAG_1 = 'tag1'
TOKEN_TEXT_1 = 'token 1'
TOKEN_TEXT_2 = 'token 2'
BOUNDING_BOX = BoundingBox(0, 0, 10, 10)
class TestAnnotateStructuredDocumentUsingPredictions(object):
def test_should_not_fail_with_empty_document(self):
structured_document = SimpleStructuredDocument()
annotate_structured_document_using_predictions(structured_document, [])
def test_should_tag_single_token_using_prediction(self):
token_1 = SimpleToken(TOKEN_TEXT_1)
structured_document = SimpleStructuredDocument(
lines=[SimpleLine([token_1])])
annotate_structured_document_using_predictions(structured_document,
[TAG_1])
assert structured_document.get_tag(token_1,
scope=CRF_TAG_SCOPE) == TAG_1
def test_should_equal_same_bounding_boxes(self):
assert BoundingBox(11, 12, 101, 102) == BoundingBox(11, 12, 101, 102)
