sebastien/Copies
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial support for "_", "|_" et "…|" in label names

Browse Source
This commit is contained in:
sebastien
2026-05-08 22:20:54 +02:00
parent 26d02b0987
commit 8d9165d0ac
4 changed files with 84 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
splitting_int.py
+70 -34
View File
@@ -27,7 +27,7 @@ def decode_json(pdf_file):
(b, label) = d["box_2d"], d["label"]
pn = page_number(b)
carreau = 1000 // 38
result.append((label, pn, b[0] - int(carreau), b[2]-int(carreau)))
result.append((label, pn, b[0] - int(carreau), b[2]-int(carreau), b[1], b[3]))
result.sort(key=lambda x: (x[1], x[2]))
return (name, result)
@@ -39,11 +39,23 @@ def split_an_interro(base_dir, input_pdf, coords_list):
generated_files = set()
parts_by_label = defaultdict(list)
# Filter consecutive duplicate labels
# 1. Parse labels to strip '|' and determine type: L (Left), R (Right), N (Normal)
parsed_coords = []
for item in coords_list:
label, pn, y0, y1, x0, x1 = item
if label.startswith("|"):
c_type, clean_label = "L", label[1:]
elif label.endswith("|"):
c_type, clean_label = "R", label[:-1]
else:
c_type, clean_label = "N", label
parsed_coords.append((clean_label, c_type, pn, y0, y1, x0, x1))
# 2. Filter consecutive duplicate labels based on the cleaned name
filtered_coords = []
if coords_list:
filtered_coords.append(coords_list[0])
for item in coords_list[1:]:
if parsed_coords:
filtered_coords.append(parsed_coords[0])
for item in parsed_coords[1:]:
if item[0] != filtered_coords[-1][0]:
filtered_coords.append(item)
coords_list = filtered_coords
@@ -53,11 +65,11 @@ def split_an_interro(base_dir, input_pdf, coords_list):
page_height = page.rect.height
return (y / 1000) * page_height
def save_cropped_page(doc, page_num, y0, y1, out_path):
def save_cropped_page(doc, page_num, x0, y0, x1, y1, out_path):
"""Saves a cropped portion of a page as a new PDF."""
page = doc[page_num]
rotated_rect = page.rect * page.transformation_matrix
visual_crop_rect = fitz.Rect(rotated_rect.x0, y0, rotated_rect.x1, y1)
visual_crop_rect = fitz.Rect(rotated_rect.x0 + x0, y0, rotated_rect.x0 + x1, y1)
unrotated_clip_rect = visual_crop_rect * page.derotation_matrix
temp_doc = fitz.open()
@@ -76,46 +88,70 @@ def split_an_interro(base_dir, input_pdf, coords_list):
temp_doc.close()
# Iterate through all labels
for idx, (title, start_page, y_start_raw, _) in enumerate(coords_list):
for idx, (clean_label, c_type, start_page, y_start_raw, y_end_box, x0_raw, x1_raw) in enumerate(coords_list):
if clean_label == "_":
continue
temp_parts = []
end_page = doc.page_count - 1
end_y_target_raw = 1000
# Determine the stopping point for this label
if idx + 1 < len(coords_list):
# Normal case: stop at the next label
_, end_page, _, y_end_raw = coords_list[idx + 1]
end_y_target_raw = y_end_raw
else:
# FIX BUG 1: Last label extends to the very end of the document
end_page = doc.page_count - 1
end_y_target_raw = 1000 # 1000 represents full height
# RULE 2: Determine stopping label
for next_item in coords_list[idx + 1:]:
n_clean, n_type, n_pn, n_y_start, _, _, _ = next_item
if c_type == "L":
is_stop = (n_type in ("L", "N"))
elif c_type == "R":
is_stop = (n_type in ("R", "N"))
else:
is_stop = True # Normal labels stop at anything
if is_stop:
end_page = n_pn
end_y_target_raw = n_y_start
break
# RULES 3 & 4: Calculate horizontal boundaries (0.0 to 1.0 fraction of local page width)
col_w = 1000 / doc.page_count
if c_type == "L": # |name
fraction_x0 = (x0_raw % col_w) / col_w
fraction_x1 = 1.0
end_y_target_raw = min(1000, end_y_target_raw + 40)
elif c_type == "R": # name|
fraction_x0 = 0.0
# Find the closest 'L' label in y-distance
L_labels = [it for it in parsed_coords if it[1] == "L"]
if L_labels:
closest_L = min(L_labels, key=lambda it: abs(it[3] - y_start_raw))
closest_L_x_center = (closest_L[5] + closest_L[6]) / 2.0
fraction_x1 = (closest_L_x_center % col_w) / col_w
if fraction_x1 <= fraction_x0: fraction_x1 = 1.0 # Fallback
else:
fraction_x1 = 1.0
else: # Normal
fraction_x0 = 0.0
fraction_x1 = 1.0
# FIX BUG 2: Iterate through EVERY page from start to end
# This handles cases where start_page == end_page, start_page + 1 == end_page,
# AND start_page + N == end_page (gaps)
current_p = start_page
while current_p <= end_page:
page = doc[current_p]
# Determine Top Cut (y0)
if current_p == start_page:
y0 = scale_coord(y_start_raw, doc[current_p])
else:
y0 = 0 # Start from top of page for intermediate/last pages
y0 = scale_coord(y_start_raw, page) if current_p == start_page else 0
y1 = scale_coord(end_y_target_raw, page) if current_p == end_page else page.rect.height
# Determine Bottom Cut (y1)
if current_p == end_page:
y1 = scale_coord(end_y_target_raw, doc[current_p])
else:
y1 = doc[current_p].rect.height # Go to bottom of intermediate pages
# Only save if the slice has height (avoid empty files)
if y1 > y0 + 1:
# Convert fractions to absolute PDF points
x0_pdf = fraction_x0 * page.rect.width
x1_pdf = fraction_x1 * page.rect.width
temp_path = f"_part_{idx}_{current_p}.pdf"
save_cropped_page(doc, current_p, y0, y1, temp_path)
save_cropped_page(doc, current_p, x0_pdf, y0, x1_pdf, y1, temp_path)
temp_parts.append(temp_path)
current_p += 1
parts_by_label[title].extend(temp_parts)
parts_by_label[clean_label].extend(temp_parts)
output_dir.mkdir(parents=True, exist_ok=True)