def create_line(self):
if len(self.clicker.points) > 2:
if self.lines:
heights = np.median(np.array(
[[line.heights[0], line.heights[1]]
for line in self.lines]),
axis=0)
else:
heights = [5, 5]
new_line = layout.TextLine(id=None,
baseline=np.array(
self.clicker.points[:-1],
dtype=np.float),
heights=heights)
new_line.polygon = linepp.baseline_to_textline(
new_line.baseline, new_line.heights)
dummy_region = layout.RegionLayout(id='dummy',
polygon=[[-1.0, -1.0],
[-1.0, -1.0],
[-1.0, -1.0]])
dummy_region.lines.append(new_line)
self.page_layout.regions.append(dummy_region)
self.lines.append(new_line)
self.clear_interaction()
self.render()
def create_region(self):
if len(self.clicker.points) > 3:
new_region = layout.RegionLayout(
id='r{}'.format(len(self.page_layout.regions) + 1),
polygon=np.array(self.clicker.points[:-1], dtype=np.float))
self.page_layout.regions.append(new_region)
self.clear_interaction()
self.render()
def create_region_from_lines(self):
if self.selected_lines:
lines = []
for l_num in self.selected_lines:
lines.append(self.lines[l_num].polygon)
poly = linepp.region_from_textlines(lines)
new_region = layout.RegionLayout(
id='r{}'.format(len(self.page_layout.regions) + 1),
polygon=np.array(poly.simplify(5).exterior.coords,
dtype=np.float))
self.page_layout.regions.append(new_region)
self.clear_interaction()
self.render()
def create_region_from_lines(self):
if self.selected_lines:
poly_points = []
max_alpha = 0
for l_num in self.selected_lines:
poly_points += self.lines[l_num].polygon.tolist()
# heuristic to estimate appropriate alpha for region polygon
x_dist = np.amax(np.diff(self.lines[l_num].baseline[:, 0]))
height = self.lines[l_num].heights[0] + self.lines[
l_num].heights[1]
max_alpha = np.amax([max_alpha, x_dist, height])
poly = region_engine.alpha_shape(
np.stack(poly_points, axis=1).T, 1.5 * max_alpha)
new_region = layout.RegionLayout(
id='r{}'.format(len(self.page_layout.regions) + 1),
polygon=np.array(poly.simplify(5).exterior.coords,
dtype=np.float))
self.page_layout.regions.append(new_region)
self.clear_interaction()
self.render()
file=sys.stderr)
height = np.median([h[0] + h[1] for h in heights_list])
print(
f"OPTIMAL DOWNAMPLING {img.shape[0]//self.downsample}:{img.shape[1]//self.downsample}",
self.downsample, height, height / self.downsample)
self.downsample = temp_downsample
return baselines_list, heights_list, textlines_list
if __name__ == '__main__':
from pero_ocr.document_ocr import layout
test_layout = layout.PageLayout(id='test')
test_p = layout.RegionLayout('r', np.zeros((4, 2)))
engine_instance = EngineLineDetectorCNN(
'/mnt/matylda1/hradis/PERO/layout_engines/baselines_prod/parsenet_multi_alldata_fix_exported'
)
image = cv2.imread(
'/mnt/matylda1/ikodym/junk/refactor_test/8e41ecc2-57ed-412a-aa4f-d945efa7c624.jpg'
)
baselines, heights, textlines = engine_instance.detect_lines(image)
for baseline, height, textline in zip(baselines, heights, textlines):
new_textline = layout.TextLine(baseline=baseline,
heights=height,
polygon=textline)
test_p.lines.append(new_textline)