def collect_table_content(table_bboxes, elems):
"""
Returns a list of elements that are contained inside
the corresponding supplied bbox.
"""
# list of table content chars
table_contents = [[] for _ in range(len(table_bboxes))]
prev_content = None
prev_bbox = None
for cid, c in enumerate(elems):
# Annotations should not fall outside alone
if isinstance(c, LTAnno):
if prev_content is not None:
prev_content.append(c)
continue
# Generally speaking table contents should be included sequentially
# and we can avoid checking all tables for elems inside
# Elements only need to intersect the bbox for table as some
# formatting of fonts may result in slightly out of bbox text
if prev_bbox is not None and intersect(prev_bbox, c.bbox):
prev_content.append(c)
continue
# Search the rest of the tables for membership when done with
# the current one
for table_id, table_bbox in enumerate(table_bboxes):
if intersect(table_bbox, c.bbox):
prev_bbox = table_bbox
prev_content = table_contents[table_id]
prev_content.append(c)
break
return table_contents
def get_alignment_features(line_bboxes, elems, font_stat):
alignment_features = []
for line_bbox in line_bboxes:
line_bbox_ordered = (line_bbox[4], line_bbox[3], line_bbox[6], line_bbox[5])
boxes = [
elem for elem in elems.mentions if intersect(line_bbox_ordered, elem.bbox)
]
boxes_segments = [
elem for elem in elems.segments if intersect(line_bbox_ordered, elem.bbox)
]
boxes_figures = [
elem for elem in elems.figures if intersect(line_bbox_ordered, elem.bbox)
]
boxes_curves = [
elem for elem in elems.curves if intersect(line_bbox_ordered, elem.bbox)
]
page_width = elems.layout.width
# page_height = elems.layout.height
avg_font_pts = get_most_common_font_pts(elems.mentions, font_stat)
width = get_page_width(boxes + boxes_segments + boxes_figures + boxes_curves)
if len(boxes) == 0:
alignment_features += [[0] * 17]
continue
char_width = get_char_width(boxes)
grid_size = avg_font_pts / 2.0
for i, m in enumerate(boxes + elems.figures):
m.id = i
m.feats = defaultdict(bool)
prefix = ""
if isinstance(m, LTTextLine) and m.font_name:
prefix = m.font_name + "-" + str(m.font_size) + "-"
m.xc = (m.x0 + m.x1) / 2.0
m.yc = (m.y0 + m.y1) / 2.0
m.feats[prefix + "x0"] = m.x0_grid = m.x0 // grid_size
m.feats[prefix + "x1"] = m.x1_grid = m.x1 // grid_size
m.feats[prefix + "xc"] = m.xc_grid = m.xc // grid_size
m.feats[prefix + "yc"] = m.yc_grid = m.yc // grid_size
nodes, nodes_features = cluster_vertically_aligned_boxes(
boxes,
elems.layout.bbox,
avg_font_pts,
width,
char_width,
boxes_segments,
boxes_curves,
boxes_figures,
page_width,
True,
)
if len(nodes_features) == 0:
alignment_features += [[0] * 17]
else:
alignment_features += [nodes_features]
return alignment_features