Skip to content

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

fix(rapidoicr): fixed issue #446 (#453)

SWHL01906462569f
diff --git a/python/demo.py b/python/demo.py
index cc8808c..693248c 100644
--- a/python/demo.py
+++ b/python/demo.py
@@ -6,7 +6,7 @@ from rapidocr import RapidOCR
engine = RapidOCR()
img_url = "https://img1.baidu.com/it/u=3619974146,1266987475&fm=253&fmt=auto&app=138&f=JPEG?w=500&h=516"
-result = engine(img_url)
+result = engine(img_url, return_word_box=True, return_single_char_box=True)
print(result)
result.vis("vis_result.jpg")
diff --git a/python/rapidocr/cal_rec_boxes/main.py b/python/rapidocr/cal_rec_boxes/main.py
index 500b387..4204afb 100644
--- a/python/rapidocr/cal_rec_boxes/main.py
+++ b/python/rapidocr/cal_rec_boxes/main.py
@@ -23,7 +23,11 @@ class CalRecBoxes:
代码借鉴自PaddlePaddle/PaddleOCR和fanqie03/char-detection"""
def __call__(
- self, imgs: List[np.ndarray], dt_boxes: List[np.ndarray], rec_res: TextRecOutput
+ self,
+ imgs: List[np.ndarray],
+ dt_boxes: List[np.ndarray],
+ rec_res: TextRecOutput,
+ return_single_char_box: bool = False,
) -> TextRecOutput:
word_results = []
for idx, (img, box) in enumerate(zip(imgs, dt_boxes)):
@@ -33,7 +37,10 @@ class CalRecBoxes:
h, w = img.shape[:2]
img_box = np.array([[0, 0], [w, 0], [w, h], [0, h]])
word_box_content_list, word_box_list, conf_list = self.cal_ocr_word_box(
- rec_res.txts[idx], img_box, rec_res.word_results[idx]
+ rec_res.txts[idx],
+ img_box,
+ rec_res.word_results[idx],
+ return_single_char_box,
)
word_box_list = self.adjust_box_overlap(copy.deepcopy(word_box_list))
direction = self.get_box_direction(box)
@@ -67,7 +74,11 @@ class CalRecBoxes:
return Direction.VERTICAL if aspect_ratio >= 1.5 else Direction.HORIZONTAL
def cal_ocr_word_box(
- self, rec_txt: str, bbox: np.ndarray, word_info: WordInfo
+ self,
+ rec_txt: str,
+ bbox: np.ndarray,
+ word_info: WordInfo,
+ return_single_char_box: bool = False,
) -> Tuple[List[str], List[List[List[float]]], List[float]]:
"""Calculate the detection frame for each word based on the results of recognition and detection of ocr
汉字坐标是单字的
@@ -87,7 +98,7 @@ class CalRecBoxes:
line_cols, char_widths, word_contents = [], [], []
for word, word_col in zip(word_info.words, word_info.word_cols):
- if is_all_en_num:
+ if is_all_en_num and not return_single_char_box:
line_cols.append(word_col)
word_contents.append("".join(word))
else:
@@ -104,7 +115,7 @@ class CalRecBoxes:
char_widths, bbox_points[0], bbox_points[2], len(rec_txt)
)
- if is_all_en_num:
+ if is_all_en_num and not return_single_char_box:
word_boxes = self.calc_en_num_box(
line_cols, avg_char_width, avg_col_width, bbox_points
)
diff --git a/python/rapidocr/cli.py b/python/rapidocr/cli.py
index ca9a8a0..fc5b6c3 100644
--- a/python/rapidocr/cli.py
+++ b/python/rapidocr/cli.py
@@ -4,7 +4,6 @@
import shutil
from pathlib import Path
-
root_dir = Path(__file__).resolve().parent
DEFAULT_CFG_PATH = root_dir / "config.yaml"
@@ -18,7 +17,7 @@ def generate_cfg(args):
def check_install(ocr_engine):
- img_url = "https://github.com/RapidAI/RapidOCR/blob/a9bb7c1f44b6e00556ada90ac588f020d7637c4b/python/tests/test_files/ch_en_num.jpg?raw=true"
+ img_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.1.0/resources/test_files/ch_en_num.jpg"
result = ocr_engine(img_url)
if result.txts is None or result.txts[0] != "正品促销":
diff --git a/python/rapidocr/config.yaml b/python/rapidocr/config.yaml
index 756e9f9..f385d27 100644
--- a/python/rapidocr/config.yaml
+++ b/python/rapidocr/config.yaml
@@ -11,6 +11,7 @@ Global:
min_side_len: 30
return_word_box: false
+ return_single_char_box: false
font_path: null
diff --git a/python/rapidocr/inference_engine/onnxruntime/provider_config.py b/python/rapidocr/inference_engine/onnxruntime/provider_config.py
index 0e58704..f276239 100644
--- a/python/rapidocr/inference_engine/onnxruntime/provider_config.py
+++ b/python/rapidocr/inference_engine/onnxruntime/provider_config.py
@@ -3,7 +3,7 @@
# @Contact: [email protected]
import platform
from enum import Enum
-from typing import Any, Dict, List, Sequence
+from typing import Any, Dict, List, Sequence, Tuple
from onnxruntime import get_available_providers, get_device
@@ -30,7 +30,7 @@ class ProviderConfig:
self.cfg = engine_cfg
- def get_ep_list(self):
+ def get_ep_list(self) -> List[Tuple[str, Dict[str, Any]]]:
results = [(EP.CPU_EP.value, self.cpu_ep_cfg())]
if self.is_cuda_available():
@@ -49,10 +49,10 @@ class ProviderConfig:
return results
def cpu_ep_cfg(self) -> Dict[str, Any]:
- return self.cfg.cpu_ep_cfg
+ return dict(self.cfg.cpu_ep_cfg)
def cuda_ep_cfg(self) -> Dict[str, Any]:
- return self.cfg.cuda_ep_cfg
+ return dict(self.cfg.cuda_ep_cfg)
def dml_ep_cfg(self) -> Dict[str, Any]:
if self.cfg.dm_ep_cfg is not None:
@@ -63,7 +63,7 @@ class ProviderConfig:
return self.cpu_ep_cfg()
def cann_ep_cfg(self) -> Dict[str, Any]:
- return self.cfg.cann_ep_cfg
+ return dict(self.cfg.cann_ep_cfg)
def verify_providers(self, session_providers: Sequence[str]):
if not session_providers:
diff --git a/python/rapidocr/main.py b/python/rapidocr/main.py
index 629a1e4..e67d442 100644
--- a/python/rapidocr/main.py
+++ b/python/rapidocr/main.py
@@ -62,7 +62,8 @@ class RapidOCR:
self.cal_rec_boxes = CalRecBoxes()
- self.return_word_box = False
+ self.return_word_box = cfg.Global.return_word_box
+ self.return_single_char_box = cfg.Global.return_single_char_box
self.cfg = cfg
@@ -73,6 +74,7 @@ class RapidOCR:
use_cls: Optional[bool] = None,
use_rec: Optional[bool] = None,
return_word_box: bool = False,
+ return_single_char_box: bool = False,
text_score: float = 0.5,
box_thresh: float = 0.5,
unclip_ratio: float = 1.6,
@@ -82,6 +84,7 @@ class RapidOCR:
use_rec = self.use_rec if use_rec is None else use_rec
self.return_word_box = return_word_box
+ self.return_single_char_box = return_single_char_box
self.text_score = text_score
self.text_det.postprocess_op.box_thresh = box_thresh
self.text_det.postprocess_op.unclip_ratio = unclip_ratio
@@ -116,7 +119,9 @@ class RapidOCR:
and det_res.boxes is not None
and all(v for v in rec_res.word_results)
):
- rec_res = self.cal_rec_boxes(img, det_res.boxes, rec_res)
+ rec_res = self.cal_rec_boxes(
+ img, det_res.boxes, rec_res, self.return_single_char_box
+ )
origin_words = []
for one_word in rec_res.word_results:
one_word_points = one_word[2]
diff --git a/python/tests/test_main.py b/python/tests/test_main.py
index 17be855..5a7ea05 100644
--- a/python/tests/test_main.py
+++ b/python/tests/test_main.py
@@ -456,3 +456,10 @@ def test_en_word_ocr(engine, img_name: str, words: str):
result = engine(img_path, return_word_box=True)
txts, _, _ = list(zip(*result.word_results))
assert txts[0] == words
+
+
+def test_en_return_single_char_box(engine):
+ img_path = tests_dir / "en.jpg"
+ result = engine(img_path, return_word_box=True, return_single_char_box=True)
+ txts, _, _ = list(zip(*result.word_results))
+ assert txts[:3] == ("3", "M", "o")