repdf 0.1.0__tar.gz → 0.1.1__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {repdf-0.1.0 → repdf-0.1.1}/PKG-INFO +7 -4
- {repdf-0.1.0 → repdf-0.1.1}/README.md +5 -2
- {repdf-0.1.0 → repdf-0.1.1}/pyproject.toml +2 -2
- {repdf-0.1.0 → repdf-0.1.1}/pyproject.toml.orig +2 -2
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/app.py +28 -2
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/icons.py +6 -1
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/ocr.py +112 -24
- repdf-0.1.1/src/repdf/ocr_languages.py +219 -0
- repdf-0.1.1/src/repdf/ocr_models.py +82 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/packaging_smoke.py +19 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/render_worker.py +13 -4
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/renderer.py +3 -1
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/strings.py +25 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/theme.py +8 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/toolbar.py +39 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/__init__.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/__main__.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/assets/check-dark.svg +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/assets/check.svg +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/assets/partial-dark.svg +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/assets/partial.svg +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/assets/repdf.ico +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/assets/repdf.png +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/core.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/entrypoint.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/images.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/navigation.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/page_view.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/pdf_navigation.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/preferences.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/preview.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/print_preview.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/printing.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/search.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/search_panel.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/widgets.py +0 -0
- {repdf-0.1.0 → repdf-0.1.1}/src/repdf/workers.py +0 -0
|
@@ -1,13 +1,13 @@
|
|
|
1
1
|
Metadata-Version: 2.3
|
|
2
2
|
Name: repdf
|
|
3
|
-
Version: 0.1.
|
|
3
|
+
Version: 0.1.1
|
|
4
4
|
Summary: RePDF (Read the PDF): a desktop PDF reader with printing, search, bookmarks, OCR and page rearrangement.
|
|
5
5
|
Requires-Dist: onnxruntime>=1.31.0,<2
|
|
6
6
|
Requires-Dist: pillow>=11.0,<13
|
|
7
7
|
Requires-Dist: pypdf[crypto]>=5.0,<7
|
|
8
8
|
Requires-Dist: pypdfium2>=4.30,<6
|
|
9
9
|
Requires-Dist: pyside6>=6.8,<7
|
|
10
|
-
Requires-Dist: rapidocr==3.10.0
|
|
10
|
+
Requires-Dist: rapidocr[rtl]==3.10.0
|
|
11
11
|
Requires-Dist: regex>=2024.11.6
|
|
12
12
|
Requires-Python: >=3.11
|
|
13
13
|
Project-URL: Homepage, https://github.com/codemee/repdf
|
|
@@ -36,7 +36,7 @@ RePDF 代表 **Read the PDF**,命令與 Python 套件名稱為 `repdf`。這
|
|
|
36
36
|
- **macOS Apple Silicon(arm64)**:下載 `repdf-<版本>-macos-arm64.dmg`,開啟後將 `RePDF.app` 拖入 Applications,再從 Applications 啟動。
|
|
37
37
|
|
|
38
38
|
- Release 同時提供 SHA256 校驗檔。
|
|
39
|
-
- 桌面套件已包含 OCR
|
|
39
|
+
- 桌面套件已包含 OCR 執行環境與內建多語言模型;其他語言可在首次選用時下載,之後離線辨識。
|
|
40
40
|
- 目前未設定 Windows 程式碼簽章或 macOS Developer ID 簽章/公證,下載後可能出現系統安全提示。
|
|
41
41
|
|
|
42
42
|
### 2. 使用 uv tool 安裝
|
|
@@ -214,7 +214,10 @@ repdf
|
|
|
214
214
|
- 輸出完整內嵌圖片,不套用頁面上的旋轉或裁切,也不包含周圍文字與向量圖形。
|
|
215
215
|
- 點到空白處會顯示提示;同一位置有多張圖片時,選取最後繪製的圖片。
|
|
216
216
|
- 掃描頁面通常會擷取整張掃描圖,無法從中直接分離照片或圖表。
|
|
217
|
-
- OCR
|
|
217
|
+
- OCR 預設支援繁體中文、簡體中文、英文、日文及多種拉丁字母語言。點「區域 OCR」旁的箭頭,可選韓文、泰文、俄文/烏克蘭文、阿拉伯文與印地文/尼泊爾文;另提供希臘文、泰米爾文與泰盧固文。
|
|
218
|
+
- 選「自動+已下載模型」可共用內建與已下載模型,逐行選擇辨識結果,不會自行下載其他模型。
|
|
219
|
+
- 語言選單以分隔線區分自動選項、已下載與未下載三組,不顯示群組標題;已下載的模型名稱後會標記「已下載」。模型下載、遺失或損壞後,分組與標記會自動更新。
|
|
220
|
+
- 額外模型會先顯示下載確認,下載完成才切換語言;取消或失敗保留原設定。語言設定會記住,並套用於區域 OCR 及損壞文字備援。辨識後請校對結果。完整操作與限制見 [OCR 說明](docs/OCR.md)。
|
|
218
221
|
|
|
219
222
|
## 語言與配色
|
|
220
223
|
|
|
@@ -19,7 +19,7 @@ RePDF 代表 **Read the PDF**,命令與 Python 套件名稱為 `repdf`。這
|
|
|
19
19
|
- **macOS Apple Silicon(arm64)**:下載 `repdf-<版本>-macos-arm64.dmg`,開啟後將 `RePDF.app` 拖入 Applications,再從 Applications 啟動。
|
|
20
20
|
|
|
21
21
|
- Release 同時提供 SHA256 校驗檔。
|
|
22
|
-
- 桌面套件已包含 OCR
|
|
22
|
+
- 桌面套件已包含 OCR 執行環境與內建多語言模型;其他語言可在首次選用時下載,之後離線辨識。
|
|
23
23
|
- 目前未設定 Windows 程式碼簽章或 macOS Developer ID 簽章/公證,下載後可能出現系統安全提示。
|
|
24
24
|
|
|
25
25
|
### 2. 使用 uv tool 安裝
|
|
@@ -197,7 +197,10 @@ repdf
|
|
|
197
197
|
- 輸出完整內嵌圖片,不套用頁面上的旋轉或裁切,也不包含周圍文字與向量圖形。
|
|
198
198
|
- 點到空白處會顯示提示;同一位置有多張圖片時,選取最後繪製的圖片。
|
|
199
199
|
- 掃描頁面通常會擷取整張掃描圖,無法從中直接分離照片或圖表。
|
|
200
|
-
- OCR
|
|
200
|
+
- OCR 預設支援繁體中文、簡體中文、英文、日文及多種拉丁字母語言。點「區域 OCR」旁的箭頭,可選韓文、泰文、俄文/烏克蘭文、阿拉伯文與印地文/尼泊爾文;另提供希臘文、泰米爾文與泰盧固文。
|
|
201
|
+
- 選「自動+已下載模型」可共用內建與已下載模型,逐行選擇辨識結果,不會自行下載其他模型。
|
|
202
|
+
- 語言選單以分隔線區分自動選項、已下載與未下載三組,不顯示群組標題;已下載的模型名稱後會標記「已下載」。模型下載、遺失或損壞後,分組與標記會自動更新。
|
|
203
|
+
- 額外模型會先顯示下載確認,下載完成才切換語言;取消或失敗保留原設定。語言設定會記住,並套用於區域 OCR 及損壞文字備援。辨識後請校對結果。完整操作與限制見 [OCR 說明](docs/OCR.md)。
|
|
201
204
|
|
|
202
205
|
## 語言與配色
|
|
203
206
|
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
[project]
|
|
2
2
|
name = "repdf"
|
|
3
|
-
version = "0.1.
|
|
3
|
+
version = "0.1.1"
|
|
4
4
|
description = "RePDF (Read the PDF): a desktop PDF reader with printing, search, bookmarks, OCR and page rearrangement."
|
|
5
5
|
readme = "README.md"
|
|
6
6
|
requires-python = ">=3.11"
|
|
@@ -10,7 +10,7 @@ dependencies = [
|
|
|
10
10
|
"pypdf[crypto]>=5.0,<7",
|
|
11
11
|
"pypdfium2>=4.30,<6",
|
|
12
12
|
"pyside6>=6.8,<7",
|
|
13
|
-
"rapidocr==3.10.0",
|
|
13
|
+
"rapidocr[rtl]==3.10.0",
|
|
14
14
|
"regex>=2024.11.6",
|
|
15
15
|
]
|
|
16
16
|
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
[project]
|
|
2
2
|
name = "repdf"
|
|
3
|
-
version = "0.1.
|
|
3
|
+
version = "0.1.1"
|
|
4
4
|
description = "RePDF (Read the PDF): a desktop PDF reader with printing, search, bookmarks, OCR and page rearrangement."
|
|
5
5
|
readme = "README.md"
|
|
6
6
|
requires-python = ">=3.11"
|
|
@@ -11,7 +11,7 @@ dependencies = [
|
|
|
11
11
|
"pypdf[crypto]>=5.0,<7",
|
|
12
12
|
"pypdfium2>=4.30,<6",
|
|
13
13
|
"pyside6>=6.8,<7",
|
|
14
|
-
"rapidocr==3.10.0",
|
|
14
|
+
"rapidocr[rtl]==3.10.0",
|
|
15
15
|
"regex>=2024.11.6",
|
|
16
16
|
]
|
|
17
17
|
|
|
@@ -10,7 +10,7 @@ from importlib.resources import files
|
|
|
10
10
|
from pathlib import Path
|
|
11
11
|
from shiboken6 import isValid
|
|
12
12
|
|
|
13
|
-
from PySide6.QtCore import QEvent, QSettings, QSize, Qt, QThreadPool, QTimer, Slot
|
|
13
|
+
from PySide6.QtCore import QEvent, QPointF, QSettings, QSize, Qt, QThreadPool, QTimer, Slot
|
|
14
14
|
from PySide6.QtGui import (
|
|
15
15
|
QCloseEvent,
|
|
16
16
|
QAction,
|
|
@@ -66,6 +66,7 @@ from repdf.search_panel import SearchController
|
|
|
66
66
|
from repdf.page_view import PageViewport, PreviewCache
|
|
67
67
|
from repdf.navigation import NavigationController
|
|
68
68
|
from repdf.printing import PrintController
|
|
69
|
+
from repdf.ocr_languages import OcrLanguageController
|
|
69
70
|
|
|
70
71
|
|
|
71
72
|
class DocumentView(QWidget):
|
|
@@ -383,6 +384,7 @@ class MainWindow(QMainWindow):
|
|
|
383
384
|
self._next_generation = 0
|
|
384
385
|
self._render_documents = {}
|
|
385
386
|
self._text_requests = {}
|
|
387
|
+
self._ocr_generations = {}
|
|
386
388
|
self._image_requests = {}
|
|
387
389
|
self._active_tasks = set()
|
|
388
390
|
self._load_batches = {}
|
|
@@ -714,8 +716,11 @@ class MainWindow(QMainWindow):
|
|
|
714
716
|
self.region_ocr_action.setCheckable(True)
|
|
715
717
|
self.region_ocr_action.setToolTip(tr('區域 OCR:框選頁面或圖片中的區域,放開滑鼠後辨識並複製文字。'))
|
|
716
718
|
self.region_ocr_action.toggled.connect(self._set_region_ocr_mode)
|
|
717
|
-
self.
|
|
719
|
+
self.ocr_languages = OcrLanguageController(self)
|
|
720
|
+
self.ocr_language_button = self.toolbar.add_action_menu_group(self.region_ocr_action, self.ocr_languages.menu)
|
|
718
721
|
self.toolbar.widgetForAction(self.region_ocr_action).setObjectName('regionOcrTool')
|
|
722
|
+
self.ocr_language_button.setToolTip(tr('OCR 辨識語言'))
|
|
723
|
+
self.ocr_language_button.setAccessibleName(tr('OCR 辨識語言'))
|
|
719
724
|
self.extract_image_action = QAction(tr('擷取圖片'), self)
|
|
720
725
|
self.extract_image_action.setCheckable(True)
|
|
721
726
|
self.extract_image_action.setEnabled(False)
|
|
@@ -808,6 +813,8 @@ class MainWindow(QMainWindow):
|
|
|
808
813
|
button.setFixedSize(34, 34)
|
|
809
814
|
|
|
810
815
|
def _refresh_toolbar_icons(self) -> None:
|
|
816
|
+
if hasattr(self, 'ocr_language_button'):
|
|
817
|
+
self.ocr_language_button.setIcon(toolbar_icon('chevron_down'))
|
|
811
818
|
for action, kind in ((self.open_action, "open"), (self.export_action, "export"),
|
|
812
819
|
(self.split_action, "split"), (self.copy_text_action, "copy_text"),
|
|
813
820
|
(self.region_ocr_action, "ocr_region"),
|
|
@@ -868,6 +875,7 @@ class MainWindow(QMainWindow):
|
|
|
868
875
|
self._update_preference_actions()
|
|
869
876
|
self.search_controller.retranslate()
|
|
870
877
|
self.navigation.retranslate()
|
|
878
|
+
self.ocr_languages.refresh()
|
|
871
879
|
|
|
872
880
|
def changeEvent(self, event: QEvent) -> None:
|
|
873
881
|
super().changeEvent(event)
|
|
@@ -930,6 +938,7 @@ class MainWindow(QMainWindow):
|
|
|
930
938
|
for label in self._fullscreen_preview.labels.values():
|
|
931
939
|
label.set_text_mode(enabled, region=region, image=image)
|
|
932
940
|
self._text_requests.clear()
|
|
941
|
+
self._ocr_generations.clear()
|
|
933
942
|
self._image_requests.clear()
|
|
934
943
|
|
|
935
944
|
def _copy_text_selection(self, doc, page_index, label, start, end, *, copy=True) -> None:
|
|
@@ -950,12 +959,14 @@ class MainWindow(QMainWindow):
|
|
|
950
959
|
if document_id == doc.document_id:
|
|
951
960
|
self._text_requests.pop(previous)
|
|
952
961
|
self._text_requests[generation] = (doc.document_id, label, copy)
|
|
962
|
+
self._ocr_generations[doc.document_id] = generation
|
|
953
963
|
self.render_service.request_text(
|
|
954
964
|
doc.source_path, page_index,
|
|
955
965
|
(start.x() / label.width(), start.y() / label.height()),
|
|
956
966
|
(end.x() / label.width(), end.y() / label.height()),
|
|
957
967
|
generation, -1_000_000 - doc.document_id,
|
|
958
968
|
allow_ocr=copy, region_ocr=region, password=doc.password,
|
|
969
|
+
ocr_profile=self.ocr_languages.profile,
|
|
959
970
|
)
|
|
960
971
|
if copy:
|
|
961
972
|
self.statusBar().showMessage(tr('正在辨識選取區域…') if region else tr('正在擷取文字…'))
|
|
@@ -1305,6 +1316,7 @@ class MainWindow(QMainWindow):
|
|
|
1305
1316
|
if document_id == doc.document_id:
|
|
1306
1317
|
self._render_documents.pop(generation)
|
|
1307
1318
|
self.documents.pop(doc.document_id, None)
|
|
1319
|
+
self._ocr_generations.pop(doc.document_id, None)
|
|
1308
1320
|
self.search_controller.remove_document(doc.document_id)
|
|
1309
1321
|
self.navigation.remove_document(doc)
|
|
1310
1322
|
self.tabs.removeTab(index)
|
|
@@ -1340,6 +1352,19 @@ class MainWindow(QMainWindow):
|
|
|
1340
1352
|
if generation in self._text_requests:
|
|
1341
1353
|
document_id, label, copy = self._text_requests[generation]
|
|
1342
1354
|
if document_id in self.documents and not self._closing:
|
|
1355
|
+
retry = payload.get('ocr_retry')
|
|
1356
|
+
if payload.get('ocr_model_missing') and copy and retry:
|
|
1357
|
+
def retry_selection(document_id=document_id, label=label, retry=retry, generation=generation):
|
|
1358
|
+
doc = self.documents.get(document_id)
|
|
1359
|
+
if (doc is None or not isValid(label) or self._closing or not label.text_mode
|
|
1360
|
+
or self._ocr_generations.get(document_id) != generation
|
|
1361
|
+
or self.ocr_languages.profile != retry['ocr_profile']
|
|
1362
|
+
or label.region_mode != (retry['type'] == 'ocr_region')):
|
|
1363
|
+
return
|
|
1364
|
+
self._copy_text_selection(doc, retry['page'], label,
|
|
1365
|
+
QPointF(retry['start'][0] * label.width(), retry['start'][1] * label.height()),
|
|
1366
|
+
QPointF(retry['end'][0] * label.width(), retry['end'][1] * label.height()))
|
|
1367
|
+
self.ocr_languages.ensure(retry['ocr_profile'], retry_selection)
|
|
1343
1368
|
if payload.get('type') == 'ocr' and copy:
|
|
1344
1369
|
self.statusBar().showMessage(tr('選取文字疑似損壞,正在進行 OCR…'))
|
|
1345
1370
|
elif payload.get('type') == 'text':
|
|
@@ -1478,6 +1503,7 @@ class MainWindow(QMainWindow):
|
|
|
1478
1503
|
|
|
1479
1504
|
def closeEvent(self, event: QCloseEvent) -> None:
|
|
1480
1505
|
self._closing = True
|
|
1506
|
+
self.ocr_languages.close()
|
|
1481
1507
|
self.print_controller.close()
|
|
1482
1508
|
self.search_controller.stop()
|
|
1483
1509
|
self._password_queue.clear()
|
|
@@ -20,7 +20,12 @@ def toolbar_icon(kind: str) -> QIcon:
|
|
|
20
20
|
painter.setPen(QPen(QApplication.palette().color(group, role),
|
|
21
21
|
1.0, Qt.PenStyle.SolidLine, Qt.PenCapStyle.RoundCap,
|
|
22
22
|
Qt.PenJoinStyle.RoundJoin))
|
|
23
|
-
if kind ==
|
|
23
|
+
if kind == 'chevron_down':
|
|
24
|
+
pen = painter.pen()
|
|
25
|
+
pen.setWidthF(2)
|
|
26
|
+
painter.setPen(pen)
|
|
27
|
+
painter.drawPolyline(QPolygonF([QPointF(6, 9), QPointF(12, 15), QPointF(18, 9)]))
|
|
28
|
+
elif kind == "more":
|
|
24
29
|
painter.setBrush(painter.pen().color())
|
|
25
30
|
for x in (5, 12, 19):
|
|
26
31
|
painter.drawEllipse(QRectF(x - 1, 11, 2, 2))
|
|
@@ -2,6 +2,7 @@
|
|
|
2
2
|
from __future__ import annotations
|
|
3
3
|
|
|
4
4
|
import hashlib
|
|
5
|
+
import math
|
|
5
6
|
import re
|
|
6
7
|
import time
|
|
7
8
|
import unicodedata
|
|
@@ -10,12 +11,17 @@ from importlib.util import find_spec
|
|
|
10
11
|
from pathlib import Path
|
|
11
12
|
|
|
12
13
|
from PIL import ImageOps
|
|
14
|
+
from repdf.ocr_models import PROFILES, installed_profiles, model_available, model_path
|
|
13
15
|
|
|
14
16
|
|
|
15
17
|
class OcrError(RuntimeError):
|
|
16
18
|
pass
|
|
17
19
|
|
|
18
20
|
|
|
21
|
+
class OcrModelMissing(OcrError):
|
|
22
|
+
pass
|
|
23
|
+
|
|
24
|
+
|
|
19
25
|
def _clean_line(text: str) -> str:
|
|
20
26
|
# Remove OCR segmentation spaces within CJK text; keep English word spaces.
|
|
21
27
|
cjk = r'\u3400-\u9fff\uf900-\ufaff\U00020000-\U000323af'
|
|
@@ -37,28 +43,39 @@ DETECTION_SHA256 = '090f04abcd9d9a7498bc4ebf677e4cb9bdce1fe4197ddb7e529f1ef44e1f
|
|
|
37
43
|
|
|
38
44
|
|
|
39
45
|
@lru_cache(maxsize=1)
|
|
40
|
-
def _engine():
|
|
46
|
+
def _engine(profile='builtin'):
|
|
41
47
|
"""Load only on first damaged selection, in the existing background process."""
|
|
42
48
|
try:
|
|
43
49
|
spec = find_spec('rapidocr')
|
|
44
50
|
if spec is None or spec.origin is None:
|
|
45
51
|
raise OcrError('RapidOCR is not installed. Run uv sync.')
|
|
46
|
-
|
|
52
|
+
if profile not in PROFILES:
|
|
53
|
+
raise OcrError('Unknown OCR language.')
|
|
54
|
+
bundled = Path(spec.origin).parent / 'models'
|
|
55
|
+
model = bundled / MODEL_NAME
|
|
56
|
+
if profile != 'builtin':
|
|
57
|
+
if not model_available(profile):
|
|
58
|
+
raise OcrModelMissing('OCR language model is missing or damaged.')
|
|
59
|
+
model = model_path(profile)
|
|
47
60
|
if not model.is_file():
|
|
48
61
|
raise OcrError('Bundled OCR model is missing. Reinstall RapidOCR 3.10.0.')
|
|
49
|
-
if hashlib.sha256(model.read_bytes()).hexdigest() != MODEL_SHA256:
|
|
62
|
+
if profile == 'builtin' and hashlib.sha256(model.read_bytes()).hexdigest() != MODEL_SHA256:
|
|
50
63
|
raise OcrError('Bundled OCR model checksum mismatch.')
|
|
51
|
-
from rapidocr import RapidOCR
|
|
52
|
-
|
|
64
|
+
from rapidocr import ModelType, OCRVersion, RapidOCR
|
|
65
|
+
params = {
|
|
53
66
|
'Global.use_det': False,
|
|
54
67
|
'Global.use_cls': False,
|
|
55
68
|
'Global.use_preprocess_img': False,
|
|
56
69
|
'Global.log_level': 'error',
|
|
57
70
|
'Rec.model_path': str(model), # Explicit local path disables model downloads.
|
|
58
|
-
'Det.model_path': str(
|
|
71
|
+
'Det.model_path': str(bundled / DETECTION_MODEL),
|
|
59
72
|
'EngineConfig.onnxruntime.intra_op_num_threads': 2,
|
|
60
73
|
'EngineConfig.onnxruntime.inter_op_num_threads': 1,
|
|
61
|
-
}
|
|
74
|
+
}
|
|
75
|
+
if profile != 'builtin':
|
|
76
|
+
params.update({'Rec.ocr_version': OCRVersion.PPOCRV5, 'Rec.model_type': ModelType.MOBILE,
|
|
77
|
+
'Rec.lang_type': profile})
|
|
78
|
+
engine = RapidOCR(params=params)
|
|
62
79
|
return engine
|
|
63
80
|
except OcrError:
|
|
64
81
|
raise
|
|
@@ -92,18 +109,13 @@ def _line_regions(rects: list[list[float]]) -> list[list[float]]:
|
|
|
92
109
|
return lines
|
|
93
110
|
|
|
94
111
|
|
|
95
|
-
def
|
|
96
|
-
"""Render selected glyphs at 400 DPI; do not OCR arbitrary/image-only regions."""
|
|
97
|
-
if not rects:
|
|
98
|
-
raise OcrError('No selected text bounds are available.')
|
|
112
|
+
def _recognize_candidates(page, lines, profile, deadline, *, tolerate_errors=False):
|
|
99
113
|
width, height = page.get_size()
|
|
100
114
|
scale = 400 / 72
|
|
101
|
-
|
|
102
|
-
|
|
103
|
-
|
|
104
|
-
engine = _engine()
|
|
115
|
+
if time.monotonic() >= deadline:
|
|
116
|
+
raise OcrError('OCR time limit reached.')
|
|
117
|
+
engine = _engine() if profile == 'builtin' else _engine(profile)
|
|
105
118
|
output = []
|
|
106
|
-
deadline = time.monotonic() + 60
|
|
107
119
|
for left, top, right, bottom in lines:
|
|
108
120
|
# Small padding preserves antialiased glyph edges without adding neighbours.
|
|
109
121
|
x0, y0 = max(0, left * width - 0.5), max(0, top * height - 0.5)
|
|
@@ -120,19 +132,94 @@ def recognize_selection(page, rects: list[list[float]]) -> str:
|
|
|
120
132
|
try:
|
|
121
133
|
result = engine(image, use_det=False, use_cls=False, use_rec=True)
|
|
122
134
|
except Exception as exc:
|
|
135
|
+
if time.monotonic() >= deadline:
|
|
136
|
+
raise OcrError('OCR time limit reached.') from exc
|
|
137
|
+
if tolerate_errors:
|
|
138
|
+
output.append(None)
|
|
139
|
+
continue
|
|
123
140
|
raise OcrError(f'RapidOCR failed: {exc}') from exc
|
|
141
|
+
if time.monotonic() >= deadline:
|
|
142
|
+
raise OcrError('OCR time limit reached.')
|
|
124
143
|
if not result.txts or not result.scores:
|
|
125
|
-
|
|
144
|
+
output.append(None)
|
|
145
|
+
continue
|
|
126
146
|
text = result.txts[0].strip()
|
|
127
|
-
|
|
147
|
+
score = min(result.scores)
|
|
148
|
+
if not text or suspicious_text(text) or not math.isfinite(score) or not .8 <= score <= 1:
|
|
149
|
+
output.append(None)
|
|
150
|
+
else:
|
|
151
|
+
output.append((_clean_line(text), score))
|
|
152
|
+
return output
|
|
153
|
+
|
|
154
|
+
|
|
155
|
+
_SCRIPT_NAMES = {'korean': ('HANGUL',), 'th': ('THAI',), 'cyrillic': ('CYRILLIC',),
|
|
156
|
+
'arabic': ('ARABIC',), 'devanagari': ('DEVANAGARI',), 'el': ('GREEK',),
|
|
157
|
+
'ta': ('TAMIL',), 'te': ('TELUGU',)}
|
|
158
|
+
|
|
159
|
+
|
|
160
|
+
def _choose_candidate(candidates):
|
|
161
|
+
"""Conservative heuristics, not calibrated cross-model probabilities."""
|
|
162
|
+
baseline = candidates.get('builtin')
|
|
163
|
+
alternatives = []
|
|
164
|
+
for profile, candidate in candidates.items():
|
|
165
|
+
if profile == 'builtin' or candidate is None:
|
|
166
|
+
continue
|
|
167
|
+
text, score = candidate
|
|
168
|
+
letters = [char for char in text if unicodedata.category(char).startswith('L')]
|
|
169
|
+
matching = sum(any(script in unicodedata.name(char, '') for script in _SCRIPT_NAMES[profile])
|
|
170
|
+
for char in letters)
|
|
171
|
+
# A specialist returning Latin gibberish must not replace multilingual text.
|
|
172
|
+
if not letters or matching / len(letters) < .6:
|
|
173
|
+
continue
|
|
174
|
+
if baseline is not None and score < .9:
|
|
175
|
+
continue
|
|
176
|
+
alternatives.append((score, text))
|
|
177
|
+
alternatives.sort(reverse=True)
|
|
178
|
+
if alternatives:
|
|
179
|
+
score, text = alternatives[0]
|
|
180
|
+
# Conflicting scripts with nearly equal evidence remain uncertain.
|
|
181
|
+
ambiguous = len(alternatives) > 1 and score - alternatives[1][0] < .03
|
|
182
|
+
if not ambiguous and (baseline is None or score + .04 > baseline[1] + .02):
|
|
183
|
+
return text
|
|
184
|
+
if baseline is not None:
|
|
185
|
+
return baseline[0]
|
|
186
|
+
raise OcrError('OCR result is empty, damaged, ambiguous or low confidence.')
|
|
187
|
+
|
|
188
|
+
|
|
189
|
+
def recognize_selection(page, rects: list[list[float]], *, profile='builtin') -> str:
|
|
190
|
+
"""Recognize each line, optionally comparing only already installed models."""
|
|
191
|
+
if not rects:
|
|
192
|
+
raise OcrError('No selected text bounds are available.')
|
|
193
|
+
lines = _line_regions(rects)
|
|
194
|
+
if len(lines) > 30:
|
|
195
|
+
raise OcrError('Please select at most 30 lines for OCR.')
|
|
196
|
+
deadline = time.monotonic() + 60
|
|
197
|
+
if profile != 'auto':
|
|
198
|
+
candidates = _recognize_candidates(page, lines, profile, deadline)
|
|
199
|
+
if any(candidate is None for candidate in candidates):
|
|
128
200
|
raise OcrError('OCR result is empty, damaged or low confidence.')
|
|
129
|
-
|
|
130
|
-
|
|
201
|
+
return '\n'.join(candidate[0] for candidate in candidates)
|
|
202
|
+
results = {'builtin': _recognize_candidates(page, lines, 'builtin', deadline, tolerate_errors=True)}
|
|
203
|
+
# Model-major order loads each engine once per request, rather than once per line.
|
|
204
|
+
# Rendering one crop at a time also avoids retaining thirty large line images.
|
|
205
|
+
for installed in installed_profiles():
|
|
206
|
+
if time.monotonic() >= deadline:
|
|
207
|
+
raise OcrError('OCR time limit reached.')
|
|
208
|
+
try:
|
|
209
|
+
results[installed] = _recognize_candidates(page, lines, installed, deadline, tolerate_errors=True)
|
|
210
|
+
except OcrError:
|
|
211
|
+
# A model removed/damaged after discovery must not trigger a download.
|
|
212
|
+
if time.monotonic() >= deadline:
|
|
213
|
+
raise OcrError('OCR time limit reached.')
|
|
214
|
+
continue
|
|
215
|
+
if time.monotonic() >= deadline:
|
|
216
|
+
raise OcrError('OCR time limit reached.')
|
|
217
|
+
return '\n'.join(_choose_candidate({name: values[index] for name, values in results.items()})
|
|
218
|
+
for index in range(len(lines)))
|
|
131
219
|
|
|
132
220
|
|
|
133
|
-
def recognize_region(page, start, end) -> dict[str, object]:
|
|
221
|
+
def recognize_region(page, start, end, *, profile='builtin') -> dict[str, object]:
|
|
134
222
|
"""OCR the displayed page rectangle, including image-only PDF content."""
|
|
135
|
-
import math
|
|
136
223
|
if len(start) != 2 or len(end) != 2:
|
|
137
224
|
raise OcrError('Invalid OCR region.')
|
|
138
225
|
values = [float(value) for value in (*start, *end)]
|
|
@@ -158,7 +245,8 @@ def recognize_region(page, start, end) -> dict[str, object]:
|
|
|
158
245
|
finally:
|
|
159
246
|
bitmap.close()
|
|
160
247
|
try:
|
|
161
|
-
|
|
248
|
+
engine = _engine() if profile in ('builtin', 'auto') else _engine(profile)
|
|
249
|
+
result = engine(image, use_det=True, use_cls=False, use_rec=False)
|
|
162
250
|
except OcrError:
|
|
163
251
|
raise
|
|
164
252
|
except Exception as exc:
|
|
@@ -173,4 +261,4 @@ def recognize_region(page, start, end) -> dict[str, object]:
|
|
|
173
261
|
x0 + max(xs) / image.width * (x1 - x0),
|
|
174
262
|
y0 + max(ys) / image.height * (y1 - y0)])
|
|
175
263
|
rects = _line_regions(rects)
|
|
176
|
-
return {'text': recognize_selection(page, rects), 'rects': rects, 'ocr_used': True}
|
|
264
|
+
return {'text': recognize_selection(page, rects, profile=profile), 'rects': rects, 'ocr_used': True}
|
|
@@ -0,0 +1,219 @@
|
|
|
1
|
+
"""Language selection and cancellable downloads for source and frozen applications."""
|
|
2
|
+
from __future__ import annotations
|
|
3
|
+
|
|
4
|
+
import hashlib
|
|
5
|
+
|
|
6
|
+
from PySide6.QtCore import QObject, QIODevice, QSaveFile, QTimer, Qt, QUrl
|
|
7
|
+
from PySide6.QtGui import QAction, QActionGroup
|
|
8
|
+
from PySide6.QtNetwork import QNetworkAccessManager, QNetworkReply, QNetworkRequest
|
|
9
|
+
from PySide6.QtWidgets import (QDialog, QDialogButtonBox, QLabel, QMenu,
|
|
10
|
+
QProgressBar, QVBoxLayout)
|
|
11
|
+
|
|
12
|
+
from repdf.ocr_models import PROFILES, installed_profiles, model_available, model_path
|
|
13
|
+
from repdf.strings import tr
|
|
14
|
+
|
|
15
|
+
|
|
16
|
+
class OcrModelDialog(QDialog):
|
|
17
|
+
def __init__(self, profile, parent=None):
|
|
18
|
+
super().__init__(parent)
|
|
19
|
+
self.profile = profile
|
|
20
|
+
self.reply = None
|
|
21
|
+
self.file = None
|
|
22
|
+
self.network = QNetworkAccessManager(self)
|
|
23
|
+
self.setWindowTitle(tr('下載 OCR 語言模型'))
|
|
24
|
+
self.setWindowModality(Qt.WindowModality.WindowModal)
|
|
25
|
+
self.resize(460, 250)
|
|
26
|
+
layout = QVBoxLayout(self)
|
|
27
|
+
self.message = QLabel(tr('「{language}」需要額外下載模型。確認後將從 RapidOCR 官方模型來源下載至使用者資料目錄;完成後可離線使用,不會上傳 PDF。',
|
|
28
|
+
language=tr(PROFILES[profile]['label'])))
|
|
29
|
+
self.message.setWordWrap(True)
|
|
30
|
+
layout.addWidget(self.message)
|
|
31
|
+
source = QLabel('https://www.modelscope.cn/models/RapidAI/RapidOCR')
|
|
32
|
+
source.setWordWrap(True)
|
|
33
|
+
source.setTextInteractionFlags(Qt.TextInteractionFlag.TextSelectableByMouse)
|
|
34
|
+
layout.addWidget(source)
|
|
35
|
+
self.status = QLabel(tr('下載完成後才會切換辨識語言;取消則保留原設定。'))
|
|
36
|
+
self.status.setWordWrap(True)
|
|
37
|
+
self.status.setTextFormat(Qt.TextFormat.PlainText)
|
|
38
|
+
layout.addWidget(self.status)
|
|
39
|
+
self.progress = QProgressBar()
|
|
40
|
+
self.progress.setRange(0, 100)
|
|
41
|
+
layout.addWidget(self.progress)
|
|
42
|
+
buttons = QDialogButtonBox(QDialogButtonBox.StandardButton.Cancel)
|
|
43
|
+
self.download_button = buttons.addButton(tr('下載並使用'), QDialogButtonBox.ButtonRole.ActionRole)
|
|
44
|
+
self.download_button.clicked.connect(self.start)
|
|
45
|
+
buttons.rejected.connect(self.reject)
|
|
46
|
+
layout.addWidget(buttons)
|
|
47
|
+
self.timer = QTimer(self)
|
|
48
|
+
self.timer.setSingleShot(True)
|
|
49
|
+
self.timer.setInterval(180_000)
|
|
50
|
+
self.timer.timeout.connect(lambda: self._fail(tr('下載逾時,請稍後重試。')))
|
|
51
|
+
|
|
52
|
+
def start(self):
|
|
53
|
+
if self.reply is not None:
|
|
54
|
+
return
|
|
55
|
+
try:
|
|
56
|
+
destination = model_path(self.profile)
|
|
57
|
+
destination.parent.mkdir(parents=True, exist_ok=True)
|
|
58
|
+
self.file = QSaveFile(str(destination))
|
|
59
|
+
if not self.file.open(QIODevice.OpenModeFlag.WriteOnly):
|
|
60
|
+
raise OSError(self.file.errorString())
|
|
61
|
+
except OSError as exc:
|
|
62
|
+
self._fail(str(exc))
|
|
63
|
+
return
|
|
64
|
+
self.digest = hashlib.sha256()
|
|
65
|
+
self.received = 0
|
|
66
|
+
self.download_button.setEnabled(False)
|
|
67
|
+
self.status.setText(tr('正在下載模型…'))
|
|
68
|
+
self.progress.setRange(0, 0)
|
|
69
|
+
request = QNetworkRequest(QUrl(PROFILES[self.profile]['url']))
|
|
70
|
+
request.setTransferTimeout(30_000)
|
|
71
|
+
self.reply = self.network.get(request)
|
|
72
|
+
self.reply.readyRead.connect(self._read)
|
|
73
|
+
self.reply.downloadProgress.connect(self._progress)
|
|
74
|
+
self.reply.finished.connect(self._finished)
|
|
75
|
+
self.timer.start()
|
|
76
|
+
|
|
77
|
+
def _progress(self, received, total):
|
|
78
|
+
if total > 0:
|
|
79
|
+
self.progress.setRange(0, 100)
|
|
80
|
+
self.progress.setValue(int(received * 100 / total))
|
|
81
|
+
|
|
82
|
+
def _read(self):
|
|
83
|
+
if self.reply is None:
|
|
84
|
+
return
|
|
85
|
+
data = bytes(self.reply.readAll())
|
|
86
|
+
self.received += len(data)
|
|
87
|
+
if self.received > 100_000_000:
|
|
88
|
+
self._fail(tr('下載檔案大小異常。'))
|
|
89
|
+
elif self.file.write(data) != len(data):
|
|
90
|
+
self._fail(self.file.errorString())
|
|
91
|
+
else:
|
|
92
|
+
self.digest.update(data)
|
|
93
|
+
|
|
94
|
+
def _finished(self):
|
|
95
|
+
if self.reply is None:
|
|
96
|
+
return
|
|
97
|
+
self._read()
|
|
98
|
+
if self.reply is None:
|
|
99
|
+
return
|
|
100
|
+
if self.reply.error() != QNetworkReply.NetworkError.NoError:
|
|
101
|
+
self._fail(self.reply.errorString())
|
|
102
|
+
return
|
|
103
|
+
if self.digest.hexdigest() != PROFILES[self.profile]['sha256']:
|
|
104
|
+
self._fail(tr('模型校驗失敗,請重新下載。'))
|
|
105
|
+
return
|
|
106
|
+
if not self.file.commit():
|
|
107
|
+
self._fail(self.file.errorString())
|
|
108
|
+
return
|
|
109
|
+
self.file = None
|
|
110
|
+
self._cleanup()
|
|
111
|
+
self.accept()
|
|
112
|
+
|
|
113
|
+
def _cleanup(self):
|
|
114
|
+
self.timer.stop()
|
|
115
|
+
reply, self.reply = self.reply, None
|
|
116
|
+
if reply is not None:
|
|
117
|
+
reply.abort()
|
|
118
|
+
reply.deleteLater()
|
|
119
|
+
if self.file is not None:
|
|
120
|
+
self.file.cancelWriting()
|
|
121
|
+
self.file = None
|
|
122
|
+
|
|
123
|
+
def _fail(self, reason):
|
|
124
|
+
self._cleanup()
|
|
125
|
+
self.progress.setRange(0, 100)
|
|
126
|
+
self.progress.setValue(0)
|
|
127
|
+
self.status.setText(tr('下載未完成:{reason}', reason=reason))
|
|
128
|
+
self.download_button.setEnabled(True)
|
|
129
|
+
|
|
130
|
+
def reject(self):
|
|
131
|
+
self._cleanup()
|
|
132
|
+
super().reject()
|
|
133
|
+
|
|
134
|
+
|
|
135
|
+
class OcrLanguageController(QObject):
|
|
136
|
+
def __init__(self, owner):
|
|
137
|
+
super().__init__(owner)
|
|
138
|
+
self.owner = owner
|
|
139
|
+
saved = owner.settings.value('ocr/language', 'builtin')
|
|
140
|
+
self.profile = saved if saved in PROFILES else 'builtin'
|
|
141
|
+
self.dialog = None
|
|
142
|
+
self.callbacks = []
|
|
143
|
+
self.menu = QMenu(owner)
|
|
144
|
+
self.group = QActionGroup(self)
|
|
145
|
+
self.actions = {}
|
|
146
|
+
self._menu_signature = None
|
|
147
|
+
for profile, info in PROFILES.items():
|
|
148
|
+
action = QAction(tr(info['label']), self)
|
|
149
|
+
action.setCheckable(True)
|
|
150
|
+
self.group.addAction(action)
|
|
151
|
+
action.triggered.connect(lambda _checked=False, profile=profile: self.choose(profile))
|
|
152
|
+
self.actions[profile] = action
|
|
153
|
+
self.menu.aboutToShow.connect(self.refresh)
|
|
154
|
+
self.refresh()
|
|
155
|
+
|
|
156
|
+
def refresh(self):
|
|
157
|
+
installed = installed_profiles()
|
|
158
|
+
signature = installed
|
|
159
|
+
if signature != self._menu_signature:
|
|
160
|
+
self.menu.clear()
|
|
161
|
+
self.menu.addAction(self.actions['builtin'])
|
|
162
|
+
self.menu.addAction(self.actions['auto'])
|
|
163
|
+
self.menu.addSeparator()
|
|
164
|
+
for profile in installed:
|
|
165
|
+
self.menu.addAction(self.actions[profile])
|
|
166
|
+
self.menu.addSeparator()
|
|
167
|
+
for profile, info in PROFILES.items():
|
|
168
|
+
if 'url' in info and profile not in installed:
|
|
169
|
+
self.menu.addAction(self.actions[profile])
|
|
170
|
+
self._menu_signature = signature
|
|
171
|
+
for profile, action in self.actions.items():
|
|
172
|
+
label = tr(PROFILES[profile]['label'])
|
|
173
|
+
action.setText(tr('{language}(已下載)', language=label) if profile in installed else label)
|
|
174
|
+
action.setChecked(profile == self.profile)
|
|
175
|
+
self.menu.setTitle(tr('OCR 辨識語言'))
|
|
176
|
+
self.owner.region_ocr_action.setToolTip(tr('區域 OCR:框選頁面或圖片中的文字。辨識語言:{language};按旁邊箭頭切換。',
|
|
177
|
+
language=tr(PROFILES[self.profile]['label'])))
|
|
178
|
+
|
|
179
|
+
def choose(self, profile):
|
|
180
|
+
def apply():
|
|
181
|
+
self.profile = profile
|
|
182
|
+
self.owner.settings.setValue('ocr/language', profile)
|
|
183
|
+
# Results from an older language must not overwrite a newer selection.
|
|
184
|
+
self.owner._text_requests.clear()
|
|
185
|
+
self.owner._ocr_generations.clear()
|
|
186
|
+
self.refresh()
|
|
187
|
+
self.ensure(profile, apply)
|
|
188
|
+
self.refresh()
|
|
189
|
+
|
|
190
|
+
def ensure(self, profile, callback):
|
|
191
|
+
if self.owner._closing:
|
|
192
|
+
return
|
|
193
|
+
if model_available(profile):
|
|
194
|
+
callback()
|
|
195
|
+
return
|
|
196
|
+
if self.dialog is not None:
|
|
197
|
+
if self.dialog.profile == profile:
|
|
198
|
+
self.callbacks.append(callback)
|
|
199
|
+
self.dialog.raise_()
|
|
200
|
+
self.dialog.activateWindow()
|
|
201
|
+
return
|
|
202
|
+
self.callbacks = [callback]
|
|
203
|
+
self.dialog = OcrModelDialog(profile, self.owner._fullscreen_preview or self.owner)
|
|
204
|
+
self.dialog.finished.connect(self._finished)
|
|
205
|
+
self.dialog.open()
|
|
206
|
+
|
|
207
|
+
def _finished(self, result):
|
|
208
|
+
dialog, self.dialog = self.dialog, None
|
|
209
|
+
callbacks, self.callbacks = self.callbacks, []
|
|
210
|
+
dialog.deleteLater()
|
|
211
|
+
if result == QDialog.DialogCode.Accepted and not self.owner._closing:
|
|
212
|
+
for callback in callbacks:
|
|
213
|
+
callback()
|
|
214
|
+
self.refresh()
|
|
215
|
+
|
|
216
|
+
def close(self):
|
|
217
|
+
self.callbacks.clear()
|
|
218
|
+
if self.dialog is not None:
|
|
219
|
+
self.dialog.reject()
|
|
@@ -0,0 +1,82 @@
|
|
|
1
|
+
"""OCR model catalog pinned to the official RapidOCR 3.10.0 manifest."""
|
|
2
|
+
from __future__ import annotations
|
|
3
|
+
|
|
4
|
+
import hashlib
|
|
5
|
+
import os
|
|
6
|
+
from pathlib import Path
|
|
7
|
+
import sys
|
|
8
|
+
|
|
9
|
+
PROFILES = {
|
|
10
|
+
'auto': {'label': '自動+已下載模型'},
|
|
11
|
+
'builtin': {'label': '自動'},
|
|
12
|
+
'korean': {
|
|
13
|
+
'label': '韓文',
|
|
14
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/korean_PP-OCRv5_rec_mobile.onnx',
|
|
15
|
+
'sha256': 'cd6e2ea50f6943ca7271eb8c56a877a5a90720b7047fe9c41a2e541a25773c9b',
|
|
16
|
+
},
|
|
17
|
+
'th': {
|
|
18
|
+
'label': '泰文',
|
|
19
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/th_PP-OCRv5_rec_mobile.onnx',
|
|
20
|
+
'sha256': 'de541dd83161c241ff426f7ecfd602a0ba77d686cf3ab9a6c255ea82fd08006e',
|
|
21
|
+
},
|
|
22
|
+
'cyrillic': {
|
|
23
|
+
'label': '俄文/烏克蘭文(西里爾字母)',
|
|
24
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/cyrillic_PP-OCRv5_rec_mobile.onnx',
|
|
25
|
+
'sha256': '90f761b4bfcce0c8c561c0cb5c887b0971d3ec01c32164bdf7374a35b0982711',
|
|
26
|
+
},
|
|
27
|
+
'arabic': {
|
|
28
|
+
'label': '阿拉伯文',
|
|
29
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/arabic_PP-OCRv5_rec_mobile.onnx',
|
|
30
|
+
'sha256': 'c1192e632d0baa9146ae5b756a0e635e3dc63c1733737ebfd1629e87144e9295',
|
|
31
|
+
},
|
|
32
|
+
'devanagari': {
|
|
33
|
+
'label': '印地文/尼泊爾文(天城文)',
|
|
34
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/devanagari_PP-OCRv5_rec_mobile.onnx',
|
|
35
|
+
'sha256': 'd6f0a906580e3fa6b324a318718f1f31f268b6ea8ef985f91c2012a37f52c91e',
|
|
36
|
+
},
|
|
37
|
+
'el': {
|
|
38
|
+
'label': '希臘文',
|
|
39
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/el_PP-OCRv5_rec_mobile.onnx',
|
|
40
|
+
'sha256': 'b4368bccd557123c702b7549fee6cd1e94b581337d1c9b65310f109131542b7f',
|
|
41
|
+
},
|
|
42
|
+
'ta': {
|
|
43
|
+
'label': '泰米爾文',
|
|
44
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/ta_PP-OCRv5_rec_mobile.onnx',
|
|
45
|
+
'sha256': 'a42448808b7dea87597336f12438935f40353f1949e8360acd9e06b4da21bfe1',
|
|
46
|
+
},
|
|
47
|
+
'te': {
|
|
48
|
+
'label': '泰盧固文',
|
|
49
|
+
'url': 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.10.0/onnx/PP-OCRv5/rec/te_PP-OCRv5_rec_mobile.onnx',
|
|
50
|
+
'sha256': 'a3690451b50028a09a3316a1274f7c05728151ea3f8fd392696397a7fefcbd92',
|
|
51
|
+
},
|
|
52
|
+
}
|
|
53
|
+
|
|
54
|
+
|
|
55
|
+
def model_directory():
|
|
56
|
+
# Never write to site-packages, a frozen bundle, or PyInstaller's extraction dir.
|
|
57
|
+
if sys.platform == 'win32':
|
|
58
|
+
base = Path(os.environ.get('LOCALAPPDATA', Path.home() / 'AppData/Local'))
|
|
59
|
+
elif sys.platform == 'darwin':
|
|
60
|
+
base = Path.home() / 'Library/Application Support'
|
|
61
|
+
else:
|
|
62
|
+
base = Path(os.environ.get('XDG_DATA_HOME', Path.home() / '.local/share'))
|
|
63
|
+
return base / 'RePDF/ocr/models'
|
|
64
|
+
|
|
65
|
+
|
|
66
|
+
def model_path(profile):
|
|
67
|
+
return model_directory() / PROFILES[profile]['url'].rsplit('/', 1)[1]
|
|
68
|
+
|
|
69
|
+
|
|
70
|
+
def model_available(profile):
|
|
71
|
+
if profile in ('auto', 'builtin'):
|
|
72
|
+
return True
|
|
73
|
+
path = model_path(profile)
|
|
74
|
+
try:
|
|
75
|
+
return hashlib.sha256(path.read_bytes()).hexdigest() == PROFILES[profile]['sha256']
|
|
76
|
+
except OSError:
|
|
77
|
+
return False
|
|
78
|
+
|
|
79
|
+
|
|
80
|
+
def installed_profiles():
|
|
81
|
+
return tuple(profile for profile, info in PROFILES.items()
|
|
82
|
+
if 'url' in info and model_available(profile))
|
|
@@ -2,6 +2,7 @@ from __future__ import annotations
|
|
|
2
2
|
|
|
3
3
|
import json
|
|
4
4
|
import os
|
|
5
|
+
import sys
|
|
5
6
|
import tempfile
|
|
6
7
|
import traceback
|
|
7
8
|
from contextlib import ExitStack
|
|
@@ -29,6 +30,10 @@ def main(report_path: str) -> int:
|
|
|
29
30
|
from repdf.strings import tr
|
|
30
31
|
from repdf.theme import ThemeController
|
|
31
32
|
from repdf.ocr import _engine, recognize_region
|
|
33
|
+
from repdf.ocr_languages import OcrModelDialog
|
|
34
|
+
from repdf.ocr_models import PROFILES, model_path
|
|
35
|
+
from PySide6.QtNetwork import QSslSocket
|
|
36
|
+
from bidi.algorithm import get_display
|
|
32
37
|
from repdf.search import compile_query
|
|
33
38
|
import pypdfium2 as pdfium
|
|
34
39
|
from PIL import Image, ImageDraw, ImageFont
|
|
@@ -44,6 +49,17 @@ def main(report_path: str) -> int:
|
|
|
44
49
|
language.close()
|
|
45
50
|
if load_app_icon().isNull():
|
|
46
51
|
raise RuntimeError("Application icon is missing")
|
|
52
|
+
if not QSslSocket.supportsSsl():
|
|
53
|
+
raise RuntimeError('HTTPS support for optional OCR model downloads is missing')
|
|
54
|
+
if get_display('abc') != 'abc' or 'korean' not in PROFILES:
|
|
55
|
+
raise RuntimeError('Optional OCR language runtime is missing')
|
|
56
|
+
frozen_root = getattr(sys, '_MEIPASS', None)
|
|
57
|
+
if frozen_root and model_path('korean').is_relative_to(Path(frozen_root)):
|
|
58
|
+
raise RuntimeError('Optional OCR models must be stored outside the bundle')
|
|
59
|
+
download_dialog = OcrModelDialog('korean')
|
|
60
|
+
if download_dialog.reply is not None:
|
|
61
|
+
raise RuntimeError('OCR downloads must require explicit confirmation')
|
|
62
|
+
download_dialog.reject()
|
|
47
63
|
if not compile_query(r'sm.ke', regular_expression=True).search('SMOKE', timeout=1):
|
|
48
64
|
raise RuntimeError('Search regular expression runtime failed')
|
|
49
65
|
ocr_image = Image.new('RGB', (500, 100), 'white')
|
|
@@ -61,12 +77,15 @@ def main(report_path: str) -> int:
|
|
|
61
77
|
ocr_page = ocr_document[0]
|
|
62
78
|
try:
|
|
63
79
|
region_result = recognize_region(ocr_page, [0, 0], [1, 1])
|
|
80
|
+
auto_result = recognize_region(ocr_page, [0, 0], [1, 1], profile='auto')
|
|
64
81
|
finally:
|
|
65
82
|
ocr_page.close()
|
|
66
83
|
finally:
|
|
67
84
|
ocr_document.close()
|
|
68
85
|
if region_result['text'] != 'Copy this text':
|
|
69
86
|
raise RuntimeError(f"Packaged region OCR failed: {region_result['text']}")
|
|
87
|
+
if auto_result['text'] != 'Copy this text':
|
|
88
|
+
raise RuntimeError(f"Packaged automatic OCR failed: {auto_result['text']}")
|
|
70
89
|
source = root / "source.pdf"
|
|
71
90
|
writer = PdfWriter()
|
|
72
91
|
for width in (200, 300):
|
|
@@ -10,7 +10,7 @@ from typing import TextIO
|
|
|
10
10
|
|
|
11
11
|
import pypdfium2 as pdfium
|
|
12
12
|
|
|
13
|
-
from repdf.ocr import OcrError, recognize_region, recognize_selection, suspicious_text
|
|
13
|
+
from repdf.ocr import OcrError, OcrModelMissing, recognize_region, recognize_selection, suspicious_text
|
|
14
14
|
from repdf.images import extract_clicked_image
|
|
15
15
|
from repdf.pdf_navigation import read_bookmarks, read_page_label, read_page_links, resolve_destination
|
|
16
16
|
|
|
@@ -98,7 +98,7 @@ def render_document(
|
|
|
98
98
|
|
|
99
99
|
def extract_selection_result(source: Path, page_index: int, start: list[float],
|
|
100
100
|
end: list[float], *, allow_ocr: bool = False,
|
|
101
|
-
on_ocr=None, password: str | None = None) -> dict[str, object]:
|
|
101
|
+
on_ocr=None, password: str | None = None, ocr_profile='builtin') -> dict[str, object]:
|
|
102
102
|
"""Return copied text and the matching glyph bounds in rendered-page coordinates."""
|
|
103
103
|
document = pdfium.PdfDocument(str(source), password=password)
|
|
104
104
|
try:
|
|
@@ -169,10 +169,12 @@ def extract_selection_result(source: Path, page_index: int, start: list[float],
|
|
|
169
169
|
if on_ocr is not None:
|
|
170
170
|
on_ocr()
|
|
171
171
|
try:
|
|
172
|
-
result['text'] = recognize_selection(page, rects)
|
|
172
|
+
result['text'] = (recognize_selection(page, rects) if ocr_profile == 'builtin'
|
|
173
|
+
else recognize_selection(page, rects, profile=ocr_profile))
|
|
173
174
|
result['ocr_used'] = True
|
|
174
175
|
except OcrError as exc:
|
|
175
176
|
result['ocr_error'] = str(exc)
|
|
177
|
+
result['ocr_model_missing'] = isinstance(exc, OcrModelMissing)
|
|
176
178
|
return result
|
|
177
179
|
finally:
|
|
178
180
|
textpage.close()
|
|
@@ -228,7 +230,8 @@ def serve(
|
|
|
228
230
|
raise ValueError('Invalid OCR page index')
|
|
229
231
|
page = document[page_index]
|
|
230
232
|
try:
|
|
231
|
-
extracted = recognize_region(page, request['start'], request['end']
|
|
233
|
+
extracted = recognize_region(page, request['start'], request['end'],
|
|
234
|
+
profile=request.get('ocr_profile', 'builtin'))
|
|
232
235
|
finally:
|
|
233
236
|
page.close()
|
|
234
237
|
finally:
|
|
@@ -239,8 +242,11 @@ def serve(
|
|
|
239
242
|
extracted = extract_selection_result(Path(request["source"]), int(request["page"]),
|
|
240
243
|
request["start"], request["end"], allow_ocr=bool(request.get('allow_ocr', False)),
|
|
241
244
|
password=request.get('password'),
|
|
245
|
+
ocr_profile=request.get('ocr_profile', 'builtin'),
|
|
242
246
|
on_ocr=lambda: print(json.dumps({'type': 'ocr', 'generation': int(request['generation'])}),
|
|
243
247
|
file=output_stream, flush=True))
|
|
248
|
+
if extracted.get('ocr_model_missing'):
|
|
249
|
+
extracted['ocr_retry'] = {key: request[key] for key in ('type', 'page', 'start', 'end', 'ocr_profile')}
|
|
244
250
|
print(json.dumps({"type": "text", "generation": int(request["generation"]),
|
|
245
251
|
**extracted}), file=output_stream, flush=True)
|
|
246
252
|
else:
|
|
@@ -263,6 +269,9 @@ def serve(
|
|
|
263
269
|
"message": str(exc),
|
|
264
270
|
"document_id": request.get('document_id') if isinstance(request, dict) else None,
|
|
265
271
|
"timeout": isinstance(exc, TimeoutError),
|
|
272
|
+
"ocr_model_missing": isinstance(exc, OcrModelMissing),
|
|
273
|
+
"ocr_retry": ({key: request[key] for key in ('type', 'page', 'start', 'end', 'ocr_profile')}
|
|
274
|
+
if isinstance(exc, OcrModelMissing) else None),
|
|
266
275
|
}), file=output_stream, flush=True)
|
|
267
276
|
finally:
|
|
268
277
|
# Release credential references before waiting for the next request.
|
|
@@ -125,11 +125,13 @@ class RenderService(QObject):
|
|
|
125
125
|
|
|
126
126
|
def request_text(self, source: Path, page: int, start: tuple[float, float],
|
|
127
127
|
end: tuple[float, float], generation: int, document_id: int,
|
|
128
|
-
*, allow_ocr: bool = False, region_ocr: bool = False, password: str | None = None
|
|
128
|
+
*, allow_ocr: bool = False, region_ocr: bool = False, password: str | None = None,
|
|
129
|
+
ocr_profile: str = 'builtin') -> None:
|
|
129
130
|
payload: dict[str, object] = {
|
|
130
131
|
"type": "ocr_region" if region_ocr else "copy_text", "source": str(source), "page": page,
|
|
131
132
|
"start": start, "end": end, "generation": generation,
|
|
132
133
|
"allow_ocr": allow_ocr,
|
|
134
|
+
"ocr_profile": ocr_profile,
|
|
133
135
|
}
|
|
134
136
|
if password is not None:
|
|
135
137
|
payload['password'] = password
|
|
@@ -5,6 +5,31 @@ from string import Formatter
|
|
|
5
5
|
|
|
6
6
|
LANGUAGE = "zh"
|
|
7
7
|
ENGLISH = {
|
|
8
|
+
"{language}(已下載)": "{language} (downloaded)",
|
|
9
|
+
"自動+已下載模型": "Automatic + downloaded models",
|
|
10
|
+
"未下載": "Not downloaded",
|
|
11
|
+
"無": "None",
|
|
12
|
+
"已下載": "Downloaded",
|
|
13
|
+
"韓文": "Korean",
|
|
14
|
+
"泰文": "Thai",
|
|
15
|
+
"俄文/烏克蘭文(西里爾字母)": "Russian / Ukrainian (Cyrillic)",
|
|
16
|
+
"阿拉伯文": "Arabic",
|
|
17
|
+
"印地文/尼泊爾文(天城文)": "Hindi / Nepali (Devanagari)",
|
|
18
|
+
"希臘文": "Greek",
|
|
19
|
+
"泰米爾文": "Tamil",
|
|
20
|
+
"泰盧固文": "Telugu",
|
|
21
|
+
"其他語言": "Other languages",
|
|
22
|
+
"OCR 辨識語言": "OCR recognition language",
|
|
23
|
+
"下載 OCR 語言模型": "Download OCR language model",
|
|
24
|
+
"「{language}」需要額外下載模型。確認後將從 RapidOCR 官方模型來源下載至使用者資料目錄;完成後可離線使用,不會上傳 PDF。": "{language} needs an additional model. Confirm to download it from RapidOCR's official model source to your user data directory. Recognition works offline afterward; no PDF is uploaded.",
|
|
25
|
+
"下載完成後才會切換辨識語言;取消則保留原設定。": "The recognition language changes only after the download succeeds. Cancel keeps your current setting.",
|
|
26
|
+
"下載並使用": "Download and use",
|
|
27
|
+
"正在下載模型…": "Downloading model…",
|
|
28
|
+
"下載逾時,請稍後重試。": "The download timed out. Please try again later.",
|
|
29
|
+
"下載檔案大小異常。": "The download exceeds the expected size limit.",
|
|
30
|
+
"模型校驗失敗,請重新下載。": "Model verification failed. Please download again.",
|
|
31
|
+
"下載未完成:{reason}": "Download incomplete: {reason}",
|
|
32
|
+
"區域 OCR:框選頁面或圖片中的文字。辨識語言:{language};按旁邊箭頭切換。": "Region OCR: select text on a page or image. Language: {language}; use the arrow to change it.",
|
|
8
33
|
"PDF 檔案": "PDF file",
|
|
9
34
|
"列印:設定範圍、印表機並即時預覽": "Print: choose pages and printer with a live preview",
|
|
10
35
|
"印表機": "Printer",
|
|
@@ -74,6 +74,8 @@ class ThemeController(QObject):
|
|
|
74
74
|
slider_handle = "#ff9d45" if dark else "#0b57d0"
|
|
75
75
|
hover = "#3b2a56" if dark else "#e7ebf0"
|
|
76
76
|
pressed = "#4d376e" if dark else "#d8e3f3"
|
|
77
|
+
menu_background = hover if dark else card
|
|
78
|
+
menu_hover = pressed if dark else hover
|
|
77
79
|
mark_suffix = "-dark" if dark else ""
|
|
78
80
|
check = files("repdf").joinpath("assets", f"check{mark_suffix}.svg").as_posix()
|
|
79
81
|
partial = files("repdf").joinpath("assets", f"partial{mark_suffix}.svg").as_posix()
|
|
@@ -85,6 +87,12 @@ class ThemeController(QObject):
|
|
|
85
87
|
QToolBar {{ background: palette(window); border: none;
|
|
86
88
|
border-bottom: 1px solid {border}; padding: 4px; spacing: 4px; }}
|
|
87
89
|
QToolBar::separator {{ background: {border}; width: 1px; margin: 5px 4px; }}
|
|
90
|
+
QMenu {{ background: {menu_background}; color: palette(window-text);
|
|
91
|
+
border: 1px solid {border}; padding: 4px; }}
|
|
92
|
+
QMenu::item {{ padding: 5px 20px 5px 24px; }}
|
|
93
|
+
QMenu::item:selected {{ background: {menu_hover}; color: palette(window-text); }}
|
|
94
|
+
QMenu::item:disabled {{ color: {disabled}; }}
|
|
95
|
+
QMenu::separator {{ height: 1px; background: {border}; margin: 4px 6px; }}
|
|
88
96
|
QTabWidget::pane {{ border: none; }}
|
|
89
97
|
QTabBar::tab {{ background: transparent; color: palette(window-text);
|
|
90
98
|
border: none; border-bottom: 2px solid transparent;
|
|
@@ -40,6 +40,8 @@ class OverflowToolBar(QToolBar):
|
|
|
40
40
|
self.overflow_button.setAccessibleName(tr("更多工具"))
|
|
41
41
|
self.overflow_menu = QMenu(self)
|
|
42
42
|
self._overflow_controls = {}
|
|
43
|
+
self._action_groups = {}
|
|
44
|
+
self._group_buttons = {}
|
|
43
45
|
self.overflow_button.setMenu(self.overflow_menu)
|
|
44
46
|
self.overflow_button.setPopupMode(QToolButton.ToolButtonPopupMode.InstantPopup)
|
|
45
47
|
self.overflow_menu.aboutToShow.connect(self._populate_overflow)
|
|
@@ -73,6 +75,37 @@ class OverflowToolBar(QToolBar):
|
|
|
73
75
|
self._overflow_controls[action] = controls
|
|
74
76
|
return action
|
|
75
77
|
|
|
78
|
+
def widgetForAction(self, action):
|
|
79
|
+
return self._group_buttons.get(action) or super().widgetForAction(action)
|
|
80
|
+
|
|
81
|
+
def add_action_menu_group(self, action, menu):
|
|
82
|
+
group = QWidget()
|
|
83
|
+
layout = QHBoxLayout(group)
|
|
84
|
+
layout.setContentsMargins(0, 0, 0, 0)
|
|
85
|
+
layout.setSpacing(2)
|
|
86
|
+
button = QToolButton(group)
|
|
87
|
+
button.setDefaultAction(action)
|
|
88
|
+
button.setFixedSize(34, 34)
|
|
89
|
+
button.setIconSize(QSize(20, 20))
|
|
90
|
+
button.setProperty('stateButton', action.isCheckable())
|
|
91
|
+
button.setAttribute(Qt.WidgetAttribute.WA_Hover, True)
|
|
92
|
+
arrow = QToolButton(group)
|
|
93
|
+
arrow.setObjectName('ocrLanguageMenu')
|
|
94
|
+
arrow.setFixedSize(14, 34)
|
|
95
|
+
arrow.setIconSize(QSize(12, 12))
|
|
96
|
+
arrow.setIcon(toolbar_icon('chevron_down'))
|
|
97
|
+
arrow.setFocusPolicy(Qt.FocusPolicy.NoFocus)
|
|
98
|
+
arrow.setMenu(menu)
|
|
99
|
+
arrow.setPopupMode(QToolButton.ToolButtonPopupMode.InstantPopup)
|
|
100
|
+
arrow.setStyleSheet('QToolButton { border: none; padding: 0; } '
|
|
101
|
+
'QToolButton::menu-indicator { image: none; }')
|
|
102
|
+
layout.addWidget(button)
|
|
103
|
+
layout.addWidget(arrow)
|
|
104
|
+
group_action = self.addWidget(group)
|
|
105
|
+
self._action_groups[group_action] = (action, menu)
|
|
106
|
+
self._group_buttons[action] = button
|
|
107
|
+
return arrow
|
|
108
|
+
|
|
76
109
|
def _control_row(self, controls):
|
|
77
110
|
row = QWidget()
|
|
78
111
|
layout = QHBoxLayout(row)
|
|
@@ -116,8 +149,14 @@ class OverflowToolBar(QToolBar):
|
|
|
116
149
|
self.overflow_menu.addSeparator()
|
|
117
150
|
elif action in self._overflow_controls:
|
|
118
151
|
self.overflow_menu.addAction(self._control_row(self._overflow_controls[action]))
|
|
152
|
+
elif action in self._action_groups:
|
|
153
|
+
primary, menu = self._action_groups[action]
|
|
154
|
+
self.overflow_menu.addAction(primary)
|
|
155
|
+
self.overflow_menu.addMenu(menu)
|
|
119
156
|
elif not isinstance(action, QWidgetAction):
|
|
120
157
|
self.overflow_menu.addAction(action)
|
|
158
|
+
if isinstance(widget, QToolButton) and widget.menu() is not None:
|
|
159
|
+
self.overflow_menu.addMenu(widget.menu())
|
|
121
160
|
actions = self.overflow_menu.actions()
|
|
122
161
|
if actions and actions[-1].isSeparator():
|
|
123
162
|
self.overflow_menu.removeAction(actions[-1])
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|