mirrors/docling
1
0
Fork 0
mirror of https://github.com/DS4SD/docling.git synced 2025-12-08 20:58:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(ocr): auto-detect rotated pages in Tesseract (#1167)

Browse Source 
* fix(ocr): tesseract support mis-oriented documents

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* fix(ocr): update missing test data

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* fix(ocr): rotate image to the natural orientation before layout prediction

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* fix(ocr): move bounding bow rotation util to orientation.py

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* fix(ocr): refactor rotation utilities

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* chore(ocr): revert layout updates

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* chore(ocr): update e2e OCR test data

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* fix(ocr): avoid to swallow tesseract errors causing orientation detection failures

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* chore(ocr): revert layout updates

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>

* chore(ocr): update e2e OCR test data

* chore(ocr): proceed to OCR without rotation when OSD fails in `TesseractOcrCliModel`

* chore(ocr): proceed to OCR without rotation when OSD fails in `TesseractOcrModel`

* chore(ocr): default `TesseractOcrCliModel._is_auto` to `False`

* fix(ocr): fix `TesseractOcrCliModel._is_auto` computation

* chore(ocr): improve logging in case of OSD failure in `TesseractOcrCliModel` and `TesseractOcrModel`

---------

Signed-off-by: Clément Doumouro <clement.doumouro@gmail.com>
This commit is contained in:
Clément Doumouro
2025-05-21 18:12:33 +02:00
committed by GitHub
parent 90875247e5
commit 45265bf8b1
96 changed files with 9864 additions and 5258 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
tests/test_e2e_ocr_conversion.py
View File

@@ -1,6 +1,6 @@
import sys
from pathlib import Path
from typing import List
from typing import List, Tuple
from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
from docling.datamodel.base_models import InputFormat
@@ -56,33 +56,35 @@ def get_converter(ocr_options: OcrOptions):
def test_e2e_conversions():
pdf_paths = get_pdf_paths()
engines: List[OcrOptions] = [
EasyOcrOptions(),
TesseractOcrOptions(),
TesseractCliOcrOptions(),
EasyOcrOptions(force_full_page_ocr=True),
TesseractOcrOptions(force_full_page_ocr=True),
TesseractOcrOptions(force_full_page_ocr=True, lang=["auto"]),
TesseractCliOcrOptions(force_full_page_ocr=True),
TesseractCliOcrOptions(force_full_page_ocr=True, lang=["auto"]),
engines: List[Tuple[OcrOptions, bool]] = [
(EasyOcrOptions(), False),
(TesseractOcrOptions(), True),
(TesseractCliOcrOptions(), True),
(EasyOcrOptions(force_full_page_ocr=True), False),
(TesseractOcrOptions(force_full_page_ocr=True), True),
(TesseractOcrOptions(force_full_page_ocr=True, lang=["auto"]), True),
(TesseractCliOcrOptions(force_full_page_ocr=True), True),
(TesseractCliOcrOptions(force_full_page_ocr=True, lang=["auto"]), True),
]
# rapidocr is only available for Python >=3.6,<3.13
if sys.version_info < (3, 13):
engines.append(RapidOcrOptions())
engines.append(RapidOcrOptions(force_full_page_ocr=True))
engines.append((RapidOcrOptions(), False))
engines.append((RapidOcrOptions(force_full_page_ocr=True), False))
# only works on mac
if "darwin" == sys.platform:
engines.append(OcrMacOptions())
engines.append(OcrMacOptions(force_full_page_ocr=True))
engines.append((OcrMacOptions(), True))
engines.append((OcrMacOptions(force_full_page_ocr=True), True))
for ocr_options in engines:
for ocr_options, supports_rotation in engines:
print(
f"Converting with ocr_engine: {ocr_options.kind}, language: {ocr_options.lang}"
)
converter = get_converter(ocr_options=ocr_options)
for pdf_path in pdf_paths:
if not supports_rotation and "rotated" in pdf_path.name:
continue
print(f"converting {pdf_path}")
doc_result: ConversionResult = converter.convert(pdf_path)