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

keep seps over artificial boundaries to improve col separation…

Browse source 
(thresholding and decoding with artificial boundary class can
 overwrite existing column separators, which in turn can contribute
 to missing column boundaries; this prioritises seps over boundaries,
 which does not impair separation of instances, as seps will separate
 text/image/etc instances just as well as artificial boundaries)
This commit is contained in:
Robert Sachunsky 2026-04-16 04:56:38 +02:00
parent 9d6ff65e1d
commit 5a27e46b22
2 changed files with 15 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

12
src/eynollah/eynollah.py
View file

@ -628,6 +628,7 @@ class Eynollah:
thresholding_for_artificial_class=False, thresholding_for_artificial_class=False,
threshold_art_class=0.1, threshold_art_class=0.1,
artificial_class=4, artificial_class=4,
separator_class=0,
): ):
self.logger.debug("enter do_prediction_new_concept (patches=%d)", patches) self.logger.debug("enter do_prediction_new_concept (patches=%d)", patches)
@ -653,7 +654,8 @@ class Eynollah:
label=artificial_class, label=artificial_class,
only=True, only=True,
skeletonize=True, skeletonize=True,
dilate=3) dilate=3,
keep=separator_class)
conf_text = resize_image(label_p_pred[:, :, 1], img_h_page, img_w_page) conf_text = resize_image(label_p_pred[:, :, 1], img_h_page, img_w_page)
return prediction, conf_text return prediction, conf_text
@ -799,7 +801,8 @@ class Eynollah:
label=artificial_class, label=artificial_class,
only=True, only=True,
skeletonize=True, skeletonize=True,
dilate=3) dilate=3,
keep=separator_class)
gc.collect() gc.collect()
return prediction, confidence return prediction, confidence
@ -1071,7 +1074,7 @@ class Eynollah:
self.logger.debug('enter textline_contours') self.logger.debug('enter textline_contours')
kwargs = dict(artificial_class=2, kwargs = dict(artificial_class=2,
n_batch_inference=3, n_batch_inference=1 if self.tables else 3, # avoid OOM
thresholding_for_artificial_class=True, thresholding_for_artificial_class=True,
threshold_art_class=self.threshold_art_class_textline) threshold_art_class=self.threshold_art_class_textline)
if use_patches: if use_patches:
@ -1131,7 +1134,8 @@ class Eynollah:
#print("inside 2 ", time.time()-t_in) #print("inside 2 ", time.time()-t_in)
kwargs = dict(n_batch_inference=1, kwargs = dict(n_batch_inference=1,
thresholding_for_artificial_class=True, thresholding_for_artificial_class=True,
threshold_art_class=self.threshold_art_class_layout) threshold_art_class=self.threshold_art_class_layout,
separator_class=3)
if num_col_classifier == 1 or num_col_classifier == 2: if num_col_classifier == 1 or num_col_classifier == 2:
if img_height_h / img_width_h > 2.5: if img_height_h / img_width_h > 2.5:
self.logger.debug("resized to %dx%d for %d cols", self.logger.debug("resized to %dx%d for %d cols",

8
src/eynollah/utils/__init__.py
View file

@ -1953,7 +1953,8 @@ def seg_mask_label(segmap:np.ndarray,
only:bool=False, only:bool=False,
label:int=2, label:int=2,
skeletonize:bool=False, skeletonize:bool=False,
dilate:int=0 dilate:int=0,
keep:int=0,
) -> None: ) -> None:
""" """
overwrite an existing segmentation map from a binary mask with a given label overwrite an existing segmentation map from a binary mask with a given label
@ -1966,11 +1967,14 @@ def seg_mask_label(segmap:np.ndarray,
only: whether to suppress the `label` outside `mask` only: whether to suppress the `label` outside `mask`
skeletonize: whether to transform the mask to its skeleton skeletonize: whether to transform the mask to its skeleton
dilate: whether to also apply dilatation after this (convolution with square kernel of given size) dilate: whether to also apply dilatation after this (convolution with square kernel of given size)
keep: if nonzero, a clas label to be kept untouched
Use this to enforce specific confidence thresholds or rules after segmentation. Use this to enforce specific confidence thresholds or rules after segmentation.
""" """
if not mask.any(): if not mask.any():
return return
if keep:
keepmask = segmap == keep
if only: if only:
segmap[segmap == label] = 0 segmap[segmap == label] = 0
if skeletonize: if skeletonize:
@ -1982,3 +1986,5 @@ def seg_mask_label(segmap:np.ndarray,
kernel = np.ones((dilate, dilate), np.uint8) kernel = np.ones((dilate, dilate), np.uint8)
mask = cv2.dilate(mask.astype(np.uint8), kernel, iterations=1) > 0 mask = cv2.dilate(mask.astype(np.uint8), kernel, iterations=1) > 0
segmap[mask] = label segmap[mask] = label
if keep:
segmap[keepmask] = keep