qurator-spk
/
sbb_pixelwise_segmentation
mirror of https://github.com/qurator-spk/sbb_pixelwise_segmentation.git
1
0
Fork 0
Code Issues Releases Wiki Activity

visuliazation layout from eynollah page-xml output

Browse Source
pull/23/head
vahidrezanezhad 2 weeks ago
parent fc75770b73
commit 3de24243fc
2 changed files with 354 additions and 9 deletions
Show Stats Download Patch File Download Diff File

51
generate_gt_for_training.py
Unescape Escape View File

@ -366,18 +366,51 @@ def visualize_textline_segmentation(dir_xml, dir_out, dir_imgs):
co_tetxlines, y_len, x_len = get_textline_contours_for_visualization(xml_file) co_tetxlines, y_len, x_len = get_textline_contours_for_visualization(xml_file)
img_total = np.zeros((y_len, x_len, 3)) added_image = visualize_image_from_contours(co_tetxlines, img)
for cont in co_tetxlines:
img_in = np.zeros((y_len, x_len, 3)) cv2.imwrite(os.path.join(dir_out, f_name+'.png'), added_image)
img_in = cv2.fillPoly(img_in, pts =[cont], color=(1,1,1))
img_total = img_total + img_in
@main.command()
@click.option(
"--dir_xml",
"-dx",
help="directory of GT page-xml files",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--dir_out",
"-do",
help="directory where plots will be written",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--dir_imgs",
"-dimg",
help="directory of images where textline segmentation will be overlayed", )
def visualize_layout_segmentation(dir_xml, dir_out, dir_imgs):
xml_files_ind = os.listdir(dir_xml)
for ind_xml in tqdm(xml_files_ind):
indexer = 0
#print(ind_xml)
#print('########################')
xml_file = os.path.join(dir_xml,ind_xml )
f_name = Path(ind_xml).stem
img_file_name_with_format = find_format_of_given_filename_in_dir(dir_imgs, f_name)
img = cv2.imread(os.path.join(dir_imgs, img_file_name_with_format))
img_total[:,:, 0][img_total[:,:, 0]>2] = 2 co_text, co_graphic, co_sep, co_img, co_table, co_noise, y_len, x_len = get_layout_contours_for_visualization(xml_file)
img_out, _ = visualize_model_output(img_total, img, task="textline")
cv2.imwrite(os.path.join(dir_out, f_name+'.png'), img_out) added_image = visualize_image_from_contours_layout(co_text['paragraph'], co_text['header'], co_text['drop-capital'], co_sep, co_img, co_text['marginalia'], img)
cv2.imwrite(os.path.join(dir_out, f_name+'.png'), added_image)
if __name__ == "__main__": if __name__ == "__main__":

312
gt_gen_utils.py
Unescape Escape View File

@ -15,6 +15,63 @@ KERNEL = np.ones((5, 5), np.uint8)
with warnings.catch_warnings(): with warnings.catch_warnings():
warnings.simplefilter("ignore") warnings.simplefilter("ignore")
def visualize_image_from_contours_layout(co_par, co_header, co_drop, co_sep, co_image, co_marginal, img):
alpha = 0.5
blank_image = np.ones( (img.shape[:]), dtype=np.uint8) * 255
col_header = (173, 216, 230)
col_drop = (0, 191, 255)
boundary_color = (143, 216, 200)#(0, 0, 255) # Dark gray for the boundary
col_par = (0, 0, 139) # Lighter gray for the filled area
col_image = (0, 100, 0)
col_sep = (255, 0, 0)
col_marginal = (106, 90, 205)
if len(co_image)>0:
cv2.drawContours(blank_image, co_image, -1, col_image, thickness=cv2.FILLED) # Fill the contour
if len(co_sep)>0:
cv2.drawContours(blank_image, co_sep, -1, col_sep, thickness=cv2.FILLED) # Fill the contour
if len(co_header)>0:
cv2.drawContours(blank_image, co_header, -1, col_header, thickness=cv2.FILLED) # Fill the contour
if len(co_par)>0:
cv2.drawContours(blank_image, co_par, -1, col_par, thickness=cv2.FILLED) # Fill the contour
cv2.drawContours(blank_image, co_par, -1, boundary_color, thickness=1) # Draw the boundary
if len(co_drop)>0:
cv2.drawContours(blank_image, co_drop, -1, col_drop, thickness=cv2.FILLED) # Fill the contour
if len(co_marginal)>0:
cv2.drawContours(blank_image, co_marginal, -1, col_marginal, thickness=cv2.FILLED) # Fill the contour
img_final =cv2.cvtColor(blank_image, cv2.COLOR_BGR2RGB)
added_image = cv2.addWeighted(img,alpha,img_final,1- alpha,0)
return added_image
def visualize_image_from_contours(contours, img):
alpha = 0.5
blank_image = np.ones( (img.shape[:]), dtype=np.uint8) * 255
boundary_color = (0, 0, 255) # Dark gray for the boundary
fill_color = (173, 216, 230) # Lighter gray for the filled area
cv2.drawContours(blank_image, contours, -1, fill_color, thickness=cv2.FILLED) # Fill the contour
cv2.drawContours(blank_image, contours, -1, boundary_color, thickness=1) # Draw the boundary
img_final =cv2.cvtColor(blank_image, cv2.COLOR_BGR2RGB)
added_image = cv2.addWeighted(img,alpha,img_final,1- alpha,0)
return added_image
def visualize_model_output(prediction, img, task): def visualize_model_output(prediction, img, task):
if task == "binarization": if task == "binarization":
@ -224,7 +281,262 @@ def get_textline_contours_for_visualization(xml_file):
break break
co_use_case.append(np.array(c_t_in)) co_use_case.append(np.array(c_t_in))
return co_use_case, y_len, x_len return co_use_case, y_len, x_len
def get_layout_contours_for_visualization(xml_file):
tree1 = ET.parse(xml_file, parser = ET.XMLParser(encoding = 'iso-8859-5'))
root1=tree1.getroot()
alltags=[elem.tag for elem in root1.iter()]
link=alltags[0].split('}')[0]+'}'
for jj in root1.iter(link+'Page'):
y_len=int(jj.attrib['imageHeight'])
x_len=int(jj.attrib['imageWidth'])
region_tags=np.unique([x for x in alltags if x.endswith('Region')])
co_text = {'drop-capital':[], "footnote":[], "footnote-continued":[], "heading":[], "signature-mark":[], "header":[], "catch-word":[], "page-number":[], "marginalia":[], "paragraph":[]}
all_defined_textregion_types = list(co_text.keys())
co_graphic = {"handwritten-annotation":[], "decoration":[], "stamp":[], "signature":[]}
all_defined_graphic_types = list(co_graphic.keys())
co_sep=[]
co_img=[]
co_table=[]
co_noise=[]
types_text = []
for tag in region_tags:
if tag.endswith('}TextRegion') or tag.endswith('}Textregion'):
for nn in root1.iter(tag):
c_t_in = {'drop-capital':[], "footnote":[], "footnote-continued":[], "heading":[], "signature-mark":[], "header":[], "catch-word":[], "page-number":[], "marginalia":[], "paragraph":[]}
sumi=0
for vv in nn.iter():
# check the format of coords
if vv.tag==link+'Coords':
coords=bool(vv.attrib)
if coords:
p_h=vv.attrib['points'].split(' ')
if "rest_as_paragraph" in types_text:
types_text_without_paragraph = [element for element in types_text if element!='rest_as_paragraph' and element!='paragraph']
if len(types_text_without_paragraph) == 0:
if "type" in nn.attrib:
c_t_in['paragraph'].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
elif len(types_text_without_paragraph) >= 1:
if "type" in nn.attrib:
if nn.attrib['type'] in types_text_without_paragraph:
c_t_in[nn.attrib['type']].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
else:
c_t_in['paragraph'].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
else:
if "type" in nn.attrib:
if nn.attrib['type'] in all_defined_textregion_types:
c_t_in[nn.attrib['type']].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
break
else:
pass
if vv.tag==link+'Point':
if "rest_as_paragraph" in types_text:
types_text_without_paragraph = [element for element in types_text if element!='rest_as_paragraph' and element!='paragraph']
if len(types_text_without_paragraph) == 0:
if "type" in nn.attrib:
c_t_in['paragraph'].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
elif len(types_text_without_paragraph) >= 1:
if "type" in nn.attrib:
if nn.attrib['type'] in types_text_without_paragraph:
c_t_in[nn.attrib['type']].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
else:
c_t_in['paragraph'].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
else:
if "type" in nn.attrib:
if nn.attrib['type'] in all_defined_textregion_types:
c_t_in[nn.attrib['type']].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
elif vv.tag!=link+'Point' and sumi>=1:
break
for element_text in list(c_t_in.keys()):
if len(c_t_in[element_text])>0:
co_text[element_text].append(np.array(c_t_in[element_text]))
if tag.endswith('}GraphicRegion') or tag.endswith('}graphicregion'):
#print('sth')
for nn in root1.iter(tag):
c_t_in_graphic = {"handwritten-annotation":[], "decoration":[], "stamp":[], "signature":[]}
sumi=0
for vv in nn.iter():
# check the format of coords
if vv.tag==link+'Coords':
coords=bool(vv.attrib)
if coords:
p_h=vv.attrib['points'].split(' ')
if "rest_as_decoration" in types_graphic:
types_graphic_without_decoration = [element for element in types_graphic if element!='rest_as_decoration' and element!='decoration']
if len(types_graphic_without_decoration) == 0:
if "type" in nn.attrib:
c_t_in_graphic['decoration'].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
elif len(types_graphic_without_decoration) >= 1:
if "type" in nn.attrib:
if nn.attrib['type'] in types_graphic_without_decoration:
c_t_in_graphic[nn.attrib['type']].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
else:
c_t_in_graphic['decoration'].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
else:
if "type" in nn.attrib:
if nn.attrib['type'] in all_defined_graphic_types:
c_t_in_graphic[nn.attrib['type']].append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
break
else:
pass
if vv.tag==link+'Point':
if "rest_as_decoration" in types_graphic:
types_graphic_without_decoration = [element for element in types_graphic if element!='rest_as_decoration' and element!='decoration']
if len(types_graphic_without_decoration) == 0:
if "type" in nn.attrib:
c_t_in_graphic['decoration'].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
elif len(types_graphic_without_decoration) >= 1:
if "type" in nn.attrib:
if nn.attrib['type'] in types_graphic_without_decoration:
c_t_in_graphic[nn.attrib['type']].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
else:
c_t_in_graphic['decoration'].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
else:
if "type" in nn.attrib:
if nn.attrib['type'] in all_defined_graphic_types:
c_t_in_graphic[nn.attrib['type']].append( [ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ] )
sumi+=1
elif vv.tag!=link+'Point' and sumi>=1:
break
for element_graphic in list(c_t_in_graphic.keys()):
if len(c_t_in_graphic[element_graphic])>0:
co_graphic[element_graphic].append(np.array(c_t_in_graphic[element_graphic]))
if tag.endswith('}ImageRegion') or tag.endswith('}imageregion'):
for nn in root1.iter(tag):
c_t_in=[]
sumi=0
for vv in nn.iter():
if vv.tag==link+'Coords':
coords=bool(vv.attrib)
if coords:
p_h=vv.attrib['points'].split(' ')
c_t_in.append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
break
else:
pass
if vv.tag==link+'Point':
c_t_in.append([ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ])
sumi+=1
elif vv.tag!=link+'Point' and sumi>=1:
break
co_img.append(np.array(c_t_in))
if tag.endswith('}SeparatorRegion') or tag.endswith('}separatorregion'):
for nn in root1.iter(tag):
c_t_in=[]
sumi=0
for vv in nn.iter():
# check the format of coords
if vv.tag==link+'Coords':
coords=bool(vv.attrib)
if coords:
p_h=vv.attrib['points'].split(' ')
c_t_in.append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
break
else:
pass
if vv.tag==link+'Point':
c_t_in.append([ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ])
sumi+=1
elif vv.tag!=link+'Point' and sumi>=1:
break
co_sep.append(np.array(c_t_in))
if tag.endswith('}TableRegion') or tag.endswith('}tableregion'):
#print('sth')
for nn in root1.iter(tag):
c_t_in=[]
sumi=0
for vv in nn.iter():
# check the format of coords
if vv.tag==link+'Coords':
coords=bool(vv.attrib)
if coords:
p_h=vv.attrib['points'].split(' ')
c_t_in.append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
break
else:
pass
if vv.tag==link+'Point':
c_t_in.append([ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ])
sumi+=1
#print(vv.tag,'in')
elif vv.tag!=link+'Point' and sumi>=1:
break
co_table.append(np.array(c_t_in))
if tag.endswith('}NoiseRegion') or tag.endswith('}noiseregion'):
#print('sth')
for nn in root1.iter(tag):
c_t_in=[]
sumi=0
for vv in nn.iter():
# check the format of coords
if vv.tag==link+'Coords':
coords=bool(vv.attrib)
if coords:
p_h=vv.attrib['points'].split(' ')
c_t_in.append( np.array( [ [ int(x.split(',')[0]) , int(x.split(',')[1]) ] for x in p_h] ) )
break
else:
pass
if vv.tag==link+'Point':
c_t_in.append([ int(float(vv.attrib['x'])) , int(float(vv.attrib['y'])) ])
sumi+=1
#print(vv.tag,'in')
elif vv.tag!=link+'Point' and sumi>=1:
break
co_noise.append(np.array(c_t_in))
return co_text, co_graphic, co_sep, co_img, co_table, co_noise, y_len, x_len
def get_images_of_ground_truth(gt_list, dir_in, output_dir, output_type, config_file, config_params, printspace, dir_images, dir_out_images): def get_images_of_ground_truth(gt_list, dir_in, output_dir, output_type, config_file, config_params, printspace, dir_images, dir_out_images):
""" """

Write Preview
Loading…
Cancel
Save