Skip to content

📄 Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.

fix: fixed issue #542 (#564)

SWHL13fb8cc8dbdf
diff --git a/python/rapidocr/utils/load_image.py b/python/rapidocr/utils/load_image.py
index 86ab953..6e7eb5d 100644
--- a/python/rapidocr/utils/load_image.py
+++ b/python/rapidocr/utils/load_image.py
@@ -8,7 +8,7 @@ from typing import Any, Union
import cv2
import numpy as np
import requests
-from PIL import Image, UnidentifiedImageError
+from PIL import Image, ImageOps, UnidentifiedImageError
from .utils import is_url
@@ -33,12 +33,14 @@ class LoadImage:
def load_img(self, img: InputType) -> np.ndarray:
if isinstance(img, (str, Path)):
- if is_url(img):
+ if is_url(str(img)):
img = Image.open(requests.get(img, stream=True, timeout=60).raw)
else:
self.verify_exist(img)
img = Image.open(img)
+ img = self.exif_transpose(img)
+
try:
img = self.img_to_ndarray(img)
except UnidentifiedImageError as e:
@@ -57,6 +59,21 @@ class LoadImage:
raise LoadImageError(f"{type(img)} is not supported!")
+ @staticmethod
+ def verify_exist(file_path: Union[str, Path]):
+ if not Path(file_path).exists():
+ raise LoadImageError(f"{file_path} does not exist.")
+
+ @staticmethod
+ def exif_transpose(img: Image.Image) -> Image.Image:
+ try:
+ img_corrected = ImageOps.exif_transpose(img)
+ if img_corrected is None:
+ return img
+ return img_corrected
+ except Exception as e:
+ return img
+
def img_to_ndarray(self, img: Image.Image) -> np.ndarray:
if img.mode == "1":
img = img.convert("L")
@@ -121,11 +138,6 @@ class LoadImage:
new_img = cv2.bitwise_not(new_img)
return new_img
- @staticmethod
- def verify_exist(file_path: Union[str, Path]):
- if not Path(file_path).exists():
- raise LoadImageError(f"{file_path} does not exist.")
-
class LoadImageError(Exception):
pass
diff --git a/python/tests/test_det_cls_rec.py b/python/tests/test_det_cls_rec.py
index 5fbfd73..27cd2ad 100644
--- a/python/tests/test_det_cls_rec.py
+++ b/python/tests/test_det_cls_rec.py
@@ -31,7 +31,7 @@ def test_only_cls(engine):
img_path = tests_dir / "text_cls.jpg"
result = engine(img_path, use_det=False, use_cls=True, use_rec=False)
assert len(result) == 1
- assert result.cls_res[0][0] == "0"
+ assert result.cls_res[0][0] == "180"
def test_only_rec(engine):
@@ -53,7 +53,7 @@ def test_cls_rec(engine):
assert result is not None
assert len(result) == 1
- assert result.txts[0] == "韩国小馆"
+ assert result.txts[0] == "怪我咯"
def test_det_cls_rec(engine):
diff --git a/python/tests/test_files/img_exif_orientation.jpg b/python/tests/test_files/img_exif_orientation.jpg
new file mode 100644
index 0000000..46c078f
Binary files /dev/null and b/python/tests/test_files/img_exif_orientation.jpg differ
diff --git a/python/tests/test_files/text_cls.jpg b/python/tests/test_files/text_cls.jpg
index fcd7e12..9fc0b79 100644
Binary files a/python/tests/test_files/text_cls.jpg and b/python/tests/test_files/text_cls.jpg differ
diff --git a/python/tests/test_input.py b/python/tests/test_input.py
index d1c58f8..cc744ec 100644
--- a/python/tests/test_input.py
+++ b/python/tests/test_input.py
@@ -14,8 +14,8 @@ sys.path.append(str(root_dir))
from rapidocr import LoadImageError, RapidOCR
-tests_dir = root_dir / "tests" / "test_files"
-img_path = tests_dir / "ch_en_num.jpg"
+test_dir = root_dir / "tests" / "test_files"
+img_path = test_dir / "ch_en_num.jpg"
@pytest.fixture()
@@ -24,6 +24,12 @@ def engine():
return engine
+def test_exif_transpose(engine):
+ img_path = test_dir / "img_exif_orientation.jpg"
+ result = engine(img_path, use_cls=False)
+ assert result.txts[0] == "我是中国人"
+
+
@mark.parametrize(
"img_name,gt",
[
@@ -38,7 +44,7 @@ def engine():
],
)
def test_transparent_img(engine, img_name: str, gt: str):
- img_path = tests_dir / img_name
+ img_path = test_dir / img_name
result = engine(img_path)
assert result.txts[0] == gt
@@ -52,7 +58,7 @@ def test_long_img(engine):
def test_full_black_img(engine):
- img_path = tests_dir / "empty_black.jpg"
+ img_path = test_dir / "empty_black.jpg"
result = engine(img_path)
assert result.img is None
assert result.boxes is None
@@ -104,7 +110,7 @@ def test_input_parameters(engine):
def test_input_three_ndim_two_channel(engine):
- img_npy = tests_dir / "two_dim_image.npy"
+ img_npy = test_dir / "two_dim_image.npy"
image_array = np.load(str(img_npy))
result = engine(image_array)
@@ -124,7 +130,7 @@ def test_input_three_ndim_one_channel(engine):
def test_mode_one_img(engine):
- img_path = tests_dir / "issue_170.png"
+ img_path = test_dir / "issue_170.png"
result = engine(img_path)
assert result.txts[0] == "TEST"
@@ -141,7 +147,7 @@ def test_mode_one_img(engine):
],
)
def test_letterbox_like(engine, img_name, gt_len, gt_first_len):
- img_path = tests_dir / img_name
+ img_path = test_dir / img_name
result = engine(img_path)
assert len(result) == gt_len