qurator-spk/eynollah
1
0
Fork 0
mirror of https://github.com/qurator-spk/eynollah.git synced 2026-05-01 03:32:00 +02:00
Code Issues Projects Releases Wiki Activity

make get_deskewed_masks() unconditional, call only when needed

Browse source
This commit is contained in:
Robert Sachunsky 2026-04-27 00:38:22 +02:00
parent 52eb4c9a0a
commit 6a92f0d49c
1 changed files with 32 additions and 37 deletions
Download patch file Download diff file Expand all files Collapse all files

69
src/eynollah/eynollah.py
View file

@ -1502,13 +1502,11 @@ class Eynollah:
label_imgs=2, label_imgs=2,
label_seps=3, label_seps=3,
): ):
"""detect page boundary and apply its mask/bbox, post-process column classifier result, optionally detect tables"""
#print(text_regions_p_1.shape, 'text_regions_p_1 shape run graphics')
#print(erosion_hurts, 'erosion_hurts')
t_in_gr = time.time() t_in_gr = time.time()
image_page, page_coord, cont_page = self.extract_page(image) image_page, page_coord, cont_page = self.extract_page(image)
#print("inside graphics 1 ", time.time() - t_in_gr)
if self.tables: if self.tables:
table_prediction, table_confidence = self.get_tables_from_model(image_page) table_prediction, table_confidence = self.get_tables_from_model(image_page)
else: else:
@ -1533,6 +1531,7 @@ class Eynollah:
textline_mask_tot_ea = textline_mask_tot_ea[box] textline_mask_tot_ea = textline_mask_tot_ea[box]
regions_confidence = regions_confidence[box] regions_confidence = regions_confidence[box]
textline_confidence = textline_confidence[box] textline_confidence = textline_confidence[box]
self.logger.debug("Cropped page is %dx%d", *text_regions_p_1.shape)
mask_images = (text_regions_p_1 == label_imgs).astype(np.uint8) mask_images = (text_regions_p_1 == label_imgs).astype(np.uint8)
mask_images = cv2.erode(mask_images, KERNEL, iterations=10) mask_images = cv2.erode(mask_images, KERNEL, iterations=10)
@ -1712,18 +1711,9 @@ class Eynollah:
text_regions_p, text_regions_p,
regions_without_separators, regions_without_separators,
): ):
if np.abs(slope_deskew) >= SLOPE_THRESHOLD: return (rotate_image(textline_mask_tot, slope_deskew),
textline_mask_tot_d = rotate_image(textline_mask_tot, slope_deskew) rotate_image(text_regions_p, slope_deskew),
text_regions_p_d = rotate_image(text_regions_p, slope_deskew) rotate_image(regions_without_separators, slope_deskew),
regions_without_separators_d = rotate_image(regions_without_separators, slope_deskew)
else:
textline_mask_tot_d = None
text_regions_p_d = None
regions_without_separators_d = None
return (
textline_mask_tot_d,
text_regions_p_d,
regions_without_separators_d,
) )
def run_boxes_order( def run_boxes_order(
@ -2199,8 +2189,7 @@ class Eynollah:
self.logger.info(f"Processing file: {writer.image_filename}") self.logger.info(f"Processing file: {writer.image_filename}")
self.logger.info("Step 1/5: Image Enhancement") self.logger.info("Step 1/5: Image Enhancement")
num_col_classifier, num_column_is_classified = \ num_col_classifier, num_column_is_classified = self.run_enhancement(image)
self.run_enhancement(image)
writer.scale_x = image['scale_x'] writer.scale_x = image['scale_x']
writer.scale_y = image['scale_y'] writer.scale_y = image['scale_y']
@ -2263,7 +2252,6 @@ class Eynollah:
self.logger.info("Job done in %.1fs", time.time() - t0) self.logger.info("Job done in %.1fs", time.time() - t0)
return return
#print("text region early -1 in %.1fs", time.time() - t0)
t1 = time.time() t1 = time.time()
self.logger.info("Step 2/5: Layout Analysis") self.logger.info("Step 2/5: Layout Analysis")
@ -2274,7 +2262,8 @@ class Eynollah:
textline_mask_tot_ea, textline_mask_tot_ea,
regions_confidence, regions_confidence,
textline_confidence) = self.get_early_layout(image, num_col_classifier) textline_confidence) = self.get_early_layout(image, num_col_classifier)
#print("text region early -2 in %.1fs", time.time() - t0) t2 = time.time()
self.logger.info("Eearly layout took %.1fs", t2 - t1)
if num_col_classifier == 1 or num_col_classifier ==2: if num_col_classifier == 1 or num_col_classifier ==2:
if num_col_classifier == 1: if num_col_classifier == 1:
@ -2289,8 +2278,9 @@ class Eynollah:
slope_deskew = self.run_deskew(textline_mask_tot_ea) slope_deskew = self.run_deskew(textline_mask_tot_ea)
if self.plotter: if self.plotter:
self.plotter.save_deskewed_image(slope_deskew, image['img'], image['name']) self.plotter.save_deskewed_image(slope_deskew, image['img'], image['name'])
#print("text region early -2,5 in %.1fs", time.time() - t0) t3 = time.time()
#self.logger.info("Textregion detection took %.1fs ", time.time() - t1t) self.logger.info("Deskewing took %.1fs", t3 - t2)
(num_col, num_col_classifier, (num_col, num_col_classifier,
page_coord, image_page, cont_page, page_coord, image_page, cont_page,
text_regions_p, table_prediction, textline_mask_tot_ea, text_regions_p, table_prediction, textline_mask_tot_ea,
@ -2335,8 +2325,6 @@ class Eynollah:
self.logger.info("Job done in %.1fs", time.time() - t0) self.logger.info("Job done in %.1fs", time.time() - t0)
return return
#print("text region early in %.1fs", time.time() - t0)
t1 = time.time()
if num_col_classifier in (1,2): if num_col_classifier in (1,2):
org_h_l_m = textline_mask_tot_ea.shape[0] org_h_l_m = textline_mask_tot_ea.shape[0]
org_w_l_m = textline_mask_tot_ea.shape[1] org_w_l_m = textline_mask_tot_ea.shape[1]
@ -2364,13 +2352,14 @@ class Eynollah:
num_col_classifier, num_col_classifier,
table_prediction) table_prediction)
(text_regions_p_d, if self.full_layout:
textline_mask_tot_ea_d, textline_mask_tot_ea_org[text_regions_p == label_drop_fl] = 0
regions_without_separators_d) = self.get_deskewed_masks( polygons_of_drop_capitals = return_contours_of_interested_region(text_regions_p,
slope_deskew, label_drop_fl,
text_regions_p, min_area=0.00003)
textline_mask_tot_ea, conf_drops = get_region_confidences(polygons_of_drop_capitals, regionsfl_confidence)
regions_without_separators) t6 = time.time()
self.logger.info("Full layout took %.1fs", t6 - t5)
min_area_mar = 0.00001 min_area_mar = 0.00001
marginal_mask = (text_regions_p == label_marg_fl).astype(np.uint8) marginal_mask = (text_regions_p == label_marg_fl).astype(np.uint8)
@ -2384,16 +2373,18 @@ class Eynollah:
conf_images = get_region_confidences(polygons_of_images, regions_confidence) conf_images = get_region_confidences(polygons_of_images, regions_confidence)
conf_tables = get_region_confidences(polygons_of_tables, table_confidence) conf_tables = get_region_confidences(polygons_of_tables, table_confidence)
if self.full_layout:
textline_mask_tot_ea_org[text_regions_p == label_drop_fl] = 0
polygons_of_drop_capitals = return_contours_of_interested_region(text_regions_p,
label_drop_fl,
min_area=0.00003)
conf_drops = get_region_confidences(polygons_of_drop_capitals, regionsfl_confidence)
polygons_of_textregions = return_contours_of_interested_region(text_regions_p, label_text, polygons_of_textregions = return_contours_of_interested_region(text_regions_p, label_text,
min_area=MIN_AREA_REGION) min_area=MIN_AREA_REGION)
if np.abs(slope_deskew) >= SLOPE_THRESHOLD and not self.reading_order_machine_based: if np.abs(slope_deskew) >= SLOPE_THRESHOLD and not self.reading_order_machine_based:
(text_regions_p_d,
textline_mask_tot_ea_d,
regions_without_separators_d) = self.get_deskewed_masks(
slope_deskew,
text_regions_p,
textline_mask_tot_ea,
regions_without_separators)
polygons_of_textregions_d = return_contours_of_interested_region(text_regions_p_d, label_text, polygons_of_textregions_d = return_contours_of_interested_region(text_regions_p_d, label_text,
min_area=MIN_AREA_REGION) min_area=MIN_AREA_REGION)
if (len(polygons_of_textregions) and if (len(polygons_of_textregions) and
@ -2421,6 +2412,8 @@ class Eynollah:
polygons_of_marginals = [] polygons_of_marginals = []
conf_textregions = conf_marginals conf_textregions = conf_marginals
conf_marginals = [] conf_marginals = []
t7 = time.time()
self.logger.info("Region contours took %.1fs", t7 - t6)
if not self.curved_line: if not self.curved_line:
self.logger.info("Mode: Light line detection") self.logger.info("Mode: Light line detection")
@ -2466,6 +2459,8 @@ class Eynollah:
all_found_textline_polygons, all_found_textline_polygons,
slopes, slopes,
conf_textregions) conf_textregions)
t8 = time.time()
self.logger.info("Line contours took %.1fs", t8 - t7)
(polygons_of_marginals_left, (polygons_of_marginals_left,
polygons_of_marginals_right, polygons_of_marginals_right,