qurator-spk/eynollah
1
0
Fork 0
mirror of https://github.com/qurator-spk/eynollah.git synced 2026-03-24 16:12:03 +01:00
Code Issues Projects Releases Wiki Activity

do_prediction*: rename identifiers for artificial class thresholding

Browse source 
- `do_prediction_new_concept` w/ patches: remove branches for
  `thresholding_for_artificial_class` (never used, wrong name)
- `do_prediction_new_concept` w/ patches: rename kwarg
  `thresholding_for_some_classes` →
  `thresholding_for_artificial_class`
- `do_prediction_new_concept`: introduce kwarg `artificial_class`
  (for baked constant 4)
- `do_prediction`: introduce kwarg `artificial_class`
  (for baked constant 2)
- `do_prediction/_new_concept`: rename kwargs
  `thresholding_for..._in_light_version` →
  `thresholding_for...`
- `do_prediction`: rename kwarg
  `threshold_art_class_textline` →
  `threshold_art_class`
- `do_prediction_new_concept`: rename kwarg
  `threshold_art_class_layout` →
  `threshold_art_class`
This commit is contained in:
Robert Sachunsky 2026-03-02 13:08:11 +01:00
parent b9cf68b51a
commit ff7dc31a68
1 changed files with 65 additions and 93 deletions
Download patch file Download diff file Expand all files Collapse all files

158
src/eynollah/eynollah.py
View file

@ -393,16 +393,16 @@ class Eynollah:
if label_p_pred[0][int(num_col - 1)] < 0.9 and img_w_new < width_early: if label_p_pred[0][int(num_col - 1)] < 0.9 and img_w_new < width_early:
img_new = np.copy(img) img_new = np.copy(img)
num_column_is_classified = False img_is_resized = False
#elif label_p_pred[0][int(num_col - 1)] < 0.8 and img_h_new >= 8000: #elif label_p_pred[0][int(num_col - 1)] < 0.8 and img_h_new >= 8000:
elif img_h_new >= 8000: elif img_h_new >= 8000:
img_new = np.copy(img) img_new = np.copy(img)
num_column_is_classified = False img_is_resized = False
else: else:
img_new = resize_image(img, img_h_new, img_w_new) img_new = resize_image(img, img_h_new, img_w_new)
num_column_is_classified = True img_is_resized = True
return img_new, num_column_is_classified return img_new, img_is_resized
def calculate_width_height_by_columns_1_2(self, img, num_col, width_early, label_p_pred): def calculate_width_height_by_columns_1_2(self, img, num_col, width_early, label_p_pred):
self.logger.debug("enter calculate_width_height_by_columns") self.logger.debug("enter calculate_width_height_by_columns")
@ -414,16 +414,16 @@ class Eynollah:
if label_p_pred[0][int(num_col - 1)] < 0.9 and img_w_new < width_early: if label_p_pred[0][int(num_col - 1)] < 0.9 and img_w_new < width_early:
img_new = np.copy(img) img_new = np.copy(img)
num_column_is_classified = False img_is_resized = False
#elif label_p_pred[0][int(num_col - 1)] < 0.8 and img_h_new >= 8000: #elif label_p_pred[0][int(num_col - 1)] < 0.8 and img_h_new >= 8000:
elif img_h_new >= 8000: elif img_h_new >= 8000:
img_new = np.copy(img) img_new = np.copy(img)
num_column_is_classified = False img_is_resized = False
else: else:
img_new = resize_image(img, img_h_new, img_w_new) img_new = resize_image(img, img_h_new, img_w_new)
num_column_is_classified = True img_is_resized = True
return img_new, num_column_is_classified return img_new, img_is_resized
def resize_image_with_column_classifier(self, is_image_enhanced, img_bin): def resize_image_with_column_classifier(self, is_image_enhanced, img_bin):
self.logger.debug("enter resize_image_with_column_classifier") self.logger.debug("enter resize_image_with_column_classifier")
@ -624,10 +624,12 @@ class Eynollah:
def do_prediction( def do_prediction(
self, patches, img, model, self, patches, img, model,
n_batch_inference=1, marginal_of_patch_percent=0.1, n_batch_inference=1, marginal_of_patch_percent=0.1,
thresholding_for_some_classes_in_light_version=False, thresholding_for_some_classes=False,
thresholding_for_artificial_class_in_light_version=False, thresholding_for_heading=False,
thresholding_for_fl_light_version=False, thresholding_for_artificial_class=False,
threshold_art_class_textline=0.1): threshold_art_class=0.1,
artificial_class=2,
):
self.logger.debug("enter do_prediction (patches=%d)", patches) self.logger.debug("enter do_prediction (patches=%d)", patches)
img_height_model = model.layers[-1].output_shape[1] img_height_model = model.layers[-1].output_shape[1]
@ -639,27 +641,19 @@ class Eynollah:
img = img / float(255.0) img = img / float(255.0)
img = resize_image(img, img_height_model, img_width_model) img = resize_image(img, img_height_model, img_width_model)
label_p_pred = model.predict(img[np.newaxis], verbose=0) label_p_pred = model.predict(img[np.newaxis], verbose=0)[0]
seg = np.argmax(label_p_pred, axis=3)[0] seg = np.argmax(label_p_pred, axis=2)
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
seg_art = label_p_pred[0,:,:,2] seg_art = label_p_pred[:, :, artificial_class]
seg_art = (seg_art >= threshold_art_class).astype(int)
seg_art[seg_art<threshold_art_class_textline] = 0 seg[skeletonize(seg_art)] = artificial_class
seg_art[seg_art>0] =1
skeleton_art = skeletonize(seg_art) if thresholding_for_heading:
skeleton_art = skeleton_art*1 seg_header = label_p_pred[:, :, 2]
seg[skeleton_art==1]=2 seg[seg_header >= 0.2] = 2
if thresholding_for_fl_light_version:
seg_header = label_p_pred[0,:,:,2]
seg_header[seg_header<0.2] = 0
seg_header[seg_header>0] =1
seg[seg_header==1]=2
seg_color = np.repeat(seg[:, :, np.newaxis], 3, axis=2) seg_color = np.repeat(seg[:, :, np.newaxis], 3, axis=2)
prediction_true = resize_image(seg_color, img_h_page, img_w_page).astype(np.uint8) prediction_true = resize_image(seg_color, img_h_page, img_w_page).astype(np.uint8)
@ -722,7 +716,7 @@ class Eynollah:
label_p_pred = model.predict(img_patch, verbose=0) label_p_pred = model.predict(img_patch, verbose=0)
seg = np.argmax(label_p_pred, axis=3) seg = np.argmax(label_p_pred, axis=3)
if thresholding_for_some_classes_in_light_version: if thresholding_for_some_classes:
seg_not_base = label_p_pred[:,:,:,4] seg_not_base = label_p_pred[:,:,:,4]
seg_not_base = (seg_not_base > 0.03).astype(int) seg_not_base = (seg_not_base > 0.03).astype(int)
@ -736,17 +730,17 @@ class Eynollah:
seg[seg_background==1]=0 seg[seg_background==1]=0
seg[(seg_line==1) & (seg==0)]=3 seg[(seg_line==1) & (seg==0)]=3
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
seg_art = label_p_pred[:,:,:,2] seg_art = label_p_pred[:, :, :, artificial_class]
seg_art = (seg_art >= threshold_art_class_textline).astype(int) seg_art = (seg_art >= threshold_art_class).astype(int)
##seg[seg_art==1]=2 ##seg[seg_art==1]=artificial_class
indexer_inside_batch = 0 indexer_inside_batch = 0
for i_batch, j_batch in zip(list_i_s, list_j_s): for i_batch, j_batch in zip(list_i_s, list_j_s):
seg_in = seg[indexer_inside_batch] seg_in = seg[indexer_inside_batch]
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
seg_in_art = seg_art[indexer_inside_batch] seg_in_art = seg_art[indexer_inside_batch]
index_y_u_in = list_y_u[indexer_inside_batch] index_y_u_in = list_y_u[indexer_inside_batch]
@ -797,7 +791,7 @@ class Eynollah:
inbox = np.index_exp[margin:-margin or None, inbox = np.index_exp[margin:-margin or None,
margin:-margin or None] margin:-margin or None]
prediction_true[where][inbox] = seg_in[inbox + (np.newaxis,)] prediction_true[where][inbox] = seg_in[inbox + (np.newaxis,)]
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
prediction_true[where][inbox + (1,)] = seg_in_art[inbox] prediction_true[where][inbox + (1,)] = seg_in_art[inbox]
indexer_inside_batch += 1 indexer_inside_batch += 1
@ -815,14 +809,14 @@ class Eynollah:
prediction_true = prediction_true.astype(np.uint8) prediction_true = prediction_true.astype(np.uint8)
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
kernel_min = np.ones((3, 3), np.uint8) kernel_min = np.ones((3, 3), np.uint8)
prediction_true[:,:,0][prediction_true[:,:,0]==2] = 0 prediction_true[:,:,0][prediction_true[:,:,0]==artificial_class] = 0
skeleton_art = skeletonize(prediction_true[:,:,1]).astype(np.uint8) skeleton_art = skeletonize(prediction_true[:,:,1]).astype(np.uint8)
skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1) skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1)
prediction_true[:,:,0][skeleton_art==1]=2 prediction_true[:,:,0][skeleton_art==1]=artificial_class
#del model #del model
gc.collect() gc.collect()
return prediction_true return prediction_true
@ -830,10 +824,10 @@ class Eynollah:
def do_prediction_new_concept( def do_prediction_new_concept(
self, patches, img, model, self, patches, img, model,
n_batch_inference=1, marginal_of_patch_percent=0.1, n_batch_inference=1, marginal_of_patch_percent=0.1,
thresholding_for_some_classes_in_light_version=False, thresholding_for_artificial_class=False,
thresholding_for_artificial_class_in_light_version=False, threshold_art_class=0.1,
threshold_art_class_textline=0.1, artificial_class=4,
threshold_art_class_layout=0.1): ):
self.logger.debug("enter do_prediction_new_concept") self.logger.debug("enter do_prediction_new_concept")
img_height_model = model.layers[-1].output_shape[1] img_height_model = model.layers[-1].output_shape[1]
@ -845,27 +839,28 @@ class Eynollah:
img = img / 255.0 img = img / 255.0
img = resize_image(img, img_height_model, img_width_model) img = resize_image(img, img_height_model, img_width_model)
label_p_pred = model.predict(img[np.newaxis], verbose=0) label_p_pred = model.predict(img[np.newaxis], verbose=0)[0]
seg = np.argmax(label_p_pred, axis=3)[0] seg = np.argmax(label_p_pred, axis=2)
seg_color = np.repeat(seg[:, :, np.newaxis], 3, axis=2) seg_color = np.repeat(seg[:, :, np.newaxis], 3, axis=2)
prediction_true = resize_image(seg_color, img_h_page, img_w_page).astype(np.uint8) prediction_true = resize_image(seg_color, img_h_page, img_w_page).astype(np.uint8)
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
kernel_min = np.ones((3, 3), np.uint8) kernel_min = np.ones((3, 3), np.uint8)
seg_art = label_p_pred[0,:,:,4] seg_art = label_p_pred[:, :, artificial_class]
seg_art = (seg_art >= threshold_art_class_layout).astype(int) seg_art = (seg_art >= threshold_art_class).astype(int)
#seg[seg_art==1]=4 #seg[seg_art==1]=4
seg_art = resize_image(seg_art, img_h_page, img_w_page).astype(np.uint8) seg_art = resize_image(seg_art, img_h_page, img_w_page).astype(np.uint8)
prediction_true[:,:,0][prediction_true[:,:,0]==4] = 0 prediction_true[:,:,0][prediction_true[:,:,0]==artificial_class] = 0
skeleton_art = skeletonize(seg_art).astype(np.uint8) skeleton_art = skeletonize(seg_art).astype(np.uint8)
skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1) skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1)
prediction_true[:,:,0][skeleton_art==1] = 4 prediction_true[:,:,0][skeleton_art==1] = artificial_class
return prediction_true , resize_image(label_p_pred[0, :, :, 1] , img_h_page, img_w_page) seg_text = resize_image(label_p_pred[:, :, 1] , img_h_page, img_w_page)
return prediction_true, seg_text
if img.shape[0] < img_height_model: if img.shape[0] < img_height_model:
img = resize_image(img, img_height_model, img.shape[1]) img = resize_image(img, img_height_model, img.shape[1])
@ -873,8 +868,6 @@ class Eynollah:
img = resize_image(img, img.shape[0], img_width_model) img = resize_image(img, img.shape[0], img_width_model)
self.logger.debug("Patch size: %sx%s", img_height_model, img_width_model) self.logger.debug("Patch size: %sx%s", img_height_model, img_width_model)
thresholding = (thresholding_for_artificial_class_in_light_version or
thresholding_for_some_classes_in_light_version)
margin = int(marginal_of_patch_percent * img_height_model) margin = int(marginal_of_patch_percent * img_height_model)
width_mid = img_width_model - 2 * margin width_mid = img_width_model - 2 * margin
height_mid = img_height_model - 2 * margin height_mid = img_height_model - 2 * margin
@ -930,27 +923,15 @@ class Eynollah:
label_p_pred = model.predict(img_patch,verbose=0) label_p_pred = model.predict(img_patch,verbose=0)
seg = np.argmax(label_p_pred, axis=3) seg = np.argmax(label_p_pred, axis=3)
if thresholding_for_some_classes_in_light_version: if thresholding_for_artificial_class:
seg_art = label_p_pred[:,:,:,4] seg_art = label_p_pred[:, :, :, artificial_class]
seg_art = (seg_art >= threshold_art_class).astype(int)
seg_art = (seg_art >= threshold_art_class_layout).astype(int)
seg_line = label_p_pred[:,:,:,3]
seg_line = (seg_line > 0.4).astype(int)
##seg[seg_art==1]=4
#seg[(seg_line==1) & (seg==0)]=3
if thresholding_for_artificial_class_in_light_version:
seg_art = label_p_pred[:,:,:,2]
seg_art = (seg_art >= threshold_art_class_textline).astype(int)
##seg[seg_art==1]=2
indexer_inside_batch = 0 indexer_inside_batch = 0
for i_batch, j_batch in zip(list_i_s, list_j_s): for i_batch, j_batch in zip(list_i_s, list_j_s):
seg_in = seg[indexer_inside_batch] seg_in = seg[indexer_inside_batch]
if thresholding: if thresholding_for_artificial_class:
seg_in_art = seg_art[indexer_inside_batch] seg_in_art = seg_art[indexer_inside_batch]
index_y_u_in = list_y_u[indexer_inside_batch] index_y_u_in = list_y_u[indexer_inside_batch]
@ -1006,7 +987,7 @@ class Eynollah:
# artificial boundary class map should be extra array # artificial boundary class map should be extra array
# rs: why does confidence_matrix only get text-label scores? # rs: why does confidence_matrix only get text-label scores?
# should be scores at final argmax # should be scores at final argmax
if thresholding: if thresholding_for_artificial_class:
prediction_true[where][inbox + (1,)] = seg_in_art[inbox] prediction_true[where][inbox + (1,)] = seg_in_art[inbox]
indexer_inside_batch += 1 indexer_inside_batch += 1
@ -1023,23 +1004,14 @@ class Eynollah:
prediction_true = prediction_true.astype(np.uint8) prediction_true = prediction_true.astype(np.uint8)
if thresholding_for_artificial_class_in_light_version: if thresholding_for_artificial_class:
kernel_min = np.ones((3, 3), np.uint8) kernel_min = np.ones((3, 3), np.uint8)
prediction_true[:,:,0][prediction_true[:,:,0]==2] = 0 prediction_true[:,:,0][prediction_true[:,:,0]==artificial_class] = 0
skeleton_art = skeletonize(prediction_true[:,:,1]).astype(np.uint8) skeleton_art = skeletonize(prediction_true[:,:,1]).astype(np.uint8)
skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1) skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1)
prediction_true[:,:,0][skeleton_art==1]=2 prediction_true[:,:,0][skeleton_art==1]=artificial_class
if thresholding_for_some_classes_in_light_version:
kernel_min = np.ones((3, 3), np.uint8)
prediction_true[:,:,0][prediction_true[:,:,0]==4] = 0
skeleton_art = skeletonize(prediction_true[:,:,1]).astype(np.uint8)
skeleton_art = cv2.dilate(skeleton_art, kernel_min, iterations=1)
prediction_true[:,:,0][skeleton_art==1]=4
gc.collect() gc.collect()
return prediction_true, confidence_matrix return prediction_true, confidence_matrix
@ -1124,11 +1096,11 @@ class Eynollah:
img_width_h = img.shape[1] img_width_h = img.shape[1]
model_region = self.model_zoo.get("region_fl") if patches else self.model_zoo.get("region_fl_np") model_region = self.model_zoo.get("region_fl") if patches else self.model_zoo.get("region_fl_np")
thresholding_for_fl_light_version = True thresholding_for_heading = True
if not patches: if not patches:
img = otsu_copy_binary(img).astype(np.uint8) img = otsu_copy_binary(img).astype(np.uint8)
prediction_regions = None prediction_regions = None
thresholding_for_fl_light_version = False thresholding_for_heading = False
elif cols: elif cols:
img = otsu_copy_binary(img).astype(np.uint8) img = otsu_copy_binary(img).astype(np.uint8)
if cols == 1: if cols == 1:
@ -1147,7 +1119,7 @@ class Eynollah:
prediction_regions = self.do_prediction(patches, img, model_region, prediction_regions = self.do_prediction(patches, img, model_region,
marginal_of_patch_percent=0.1, marginal_of_patch_percent=0.1,
n_batch_inference=3, n_batch_inference=3,
thresholding_for_fl_light_version=thresholding_for_fl_light_version) thresholding_for_heading=thresholding_for_heading)
prediction_regions = resize_image(prediction_regions, img_height_h, img_width_h) prediction_regions = resize_image(prediction_regions, img_height_h, img_width_h)
self.logger.debug("exit extract_text_regions") self.logger.debug("exit extract_text_regions")
return prediction_regions, prediction_regions return prediction_regions, prediction_regions
@ -1305,7 +1277,7 @@ class Eynollah:
prediction_textline = self.do_prediction(use_patches, img, self.model_zoo.get("textline"), prediction_textline = self.do_prediction(use_patches, img, self.model_zoo.get("textline"),
marginal_of_patch_percent=0.15, marginal_of_patch_percent=0.15,
n_batch_inference=3, n_batch_inference=3,
threshold_art_class_textline=self.threshold_art_class_textline) threshold_art_class=self.threshold_art_class_textline)
prediction_textline = resize_image(prediction_textline, img_h, img_w) prediction_textline = resize_image(prediction_textline, img_h, img_w)
textline_mask_tot_ea_art = (prediction_textline[:,:]==2)*1 textline_mask_tot_ea_art = (prediction_textline[:,:]==2)*1
@ -1399,15 +1371,15 @@ class Eynollah:
img_resized.shape[1], img_resized.shape[0], num_col_classifier) img_resized.shape[1], img_resized.shape[0], num_col_classifier)
prediction_regions_org, confidence_matrix = self.do_prediction_new_concept( prediction_regions_org, confidence_matrix = self.do_prediction_new_concept(
True, img_resized, self.model_zoo.get("region_1_2"), n_batch_inference=1, True, img_resized, self.model_zoo.get("region_1_2"), n_batch_inference=1,
thresholding_for_some_classes_in_light_version=True, thresholding_for_artificial_class=True,
threshold_art_class_layout=self.threshold_art_class_layout) threshold_art_class=self.threshold_art_class_layout)
else: else:
prediction_regions_org = np.zeros((self.image_org.shape[0], self.image_org.shape[1], 3)) prediction_regions_org = np.zeros((self.image_org.shape[0], self.image_org.shape[1], 3))
confidence_matrix = np.zeros((self.image_org.shape[0], self.image_org.shape[1])) confidence_matrix = np.zeros((self.image_org.shape[0], self.image_org.shape[1]))
prediction_regions_page, confidence_matrix_page = self.do_prediction_new_concept( prediction_regions_page, confidence_matrix_page = self.do_prediction_new_concept(
False, self.image_page_org_size, self.model_zoo.get("region_1_2"), n_batch_inference=1, False, self.image_page_org_size, self.model_zoo.get("region_1_2"), n_batch_inference=1,
thresholding_for_artificial_class_in_light_version=True, thresholding_for_artificial_class=True,
threshold_art_class_layout=self.threshold_art_class_layout) threshold_art_class=self.threshold_art_class_layout)
ys = slice(*self.page_coord[0:2]) ys = slice(*self.page_coord[0:2])
xs = slice(*self.page_coord[2:4]) xs = slice(*self.page_coord[2:4])
prediction_regions_org[ys, xs] = prediction_regions_page prediction_regions_org[ys, xs] = prediction_regions_page
@ -1420,11 +1392,11 @@ class Eynollah:
img_resized.shape[1], img_resized.shape[0], new_h, num_col_classifier) img_resized.shape[1], img_resized.shape[0], new_h, num_col_classifier)
prediction_regions_org, confidence_matrix = self.do_prediction_new_concept( prediction_regions_org, confidence_matrix = self.do_prediction_new_concept(
True, img_resized, self.model_zoo.get("region_1_2"), n_batch_inference=2, True, img_resized, self.model_zoo.get("region_1_2"), n_batch_inference=2,
thresholding_for_some_classes_in_light_version=True, thresholding_for_artificial_class=True,
threshold_art_class_layout=self.threshold_art_class_layout) threshold_art_class=self.threshold_art_class_layout)
###prediction_regions_org = self.do_prediction(True, img_bin, self.model_zoo.get_model("region"), ###prediction_regions_org = self.do_prediction(True, img_bin, self.model_zoo.get_model("region"),
###n_batch_inference=3, ###n_batch_inference=3,
###thresholding_for_some_classes_in_light_version=True) ###thresholding_for_some_classes=True)
#print("inside 3 ", time.time()-t_in) #print("inside 3 ", time.time()-t_in)
#plt.imshow(prediction_regions_org[:,:,0]) #plt.imshow(prediction_regions_org[:,:,0])
#plt.show() #plt.show()