Skip to content

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

Unify the rapidocr_onnxruntime between ocrweb and python directory.

SWHLcf61ec3adb32
diff --git a/README.md b/README.md
index e3edf56..ee32b4b 100755
--- a/README.md
+++ b/README.md
@@ -71,7 +71,7 @@
#### 🤖2022-12-14 update:
- \[python\] 将配置参数和模型移到模块里面,同时将模型打到whl包内,可以直接pip安装使用,更加方便快捷。
- 详情参见:[README](./python/README.md#推荐pip安装快速使用)
-- 优化ocrweb部分代码
+- 优化ocrweb部分代码,统一ocrweb中`rapidocr_onnxruntime`包与`python`目录下的`rapidocr_onnxruntime`为一个
#### 🧻2022-11-20 update:
- \[python\] 添加版面分析部分,支持中文、英文和表格三种版面的检测分析。详情参见:[Rapid Structure](./python/rapid_structure/README.md)部分。
diff --git a/docs/doc_whl_rapidocr_ort.md b/docs/doc_whl_rapidocr_ort.md
index bec8cd7..d9d730c 100644
--- a/docs/doc_whl_rapidocr_ort.md
+++ b/docs/doc_whl_rapidocr_ort.md
@@ -18,9 +18,13 @@ text_sys = RapidOCR()
img = cv2.imread('test_images/ch_en_num.jpg')
-result = text_sys(img)
+result, elapse_list = text_sys(img)
print(result)
+print(elapse_list)
# result: [[dt_boxes], txt, score]
# 示例:[[左上, 右上, 右下, 左下], '小明', '0.99']
+
+# elapse_list: [det_elapse, cls_elapse, rec_elapse]
+# all_elapse = det_elapse + cls_elapse + rec_elapse
```
diff --git a/docs/doc_whl_rapidocr_vino.md b/docs/doc_whl_rapidocr_vino.md
index e19ad75..aae98f1 100644
--- a/docs/doc_whl_rapidocr_vino.md
+++ b/docs/doc_whl_rapidocr_vino.md
@@ -15,13 +15,17 @@ $ pip install rapidocr-openvino
import cv2
from rapidocr_openvino import RapidOCR
-text_sys = RapidOCR('config.yaml')
+text_sys = RapidOCR()
img = cv2.imread('test_images/ch_en_num.jpg')
-result = text_sys(img)
+result, elapse_list = text_sys(img)
print(result)
+print(elapse_list)
# result: [[dt_boxes], txt, score]
# 示例:[[左上, 右上, 右下, 左下], '小明', '0.99']
+
+# elapse_list: [det_elapse, cls_elapse, rec_elapse]
+# all_elapse = det_elapse + cls_elapse + rec_elapse
```
diff --git a/ocrweb/rapidocr_onnxruntime/__init__.py b/ocrweb/rapidocr_onnxruntime/__init__.py
index 5a34512..6522633 100644
--- a/ocrweb/rapidocr_onnxruntime/__init__.py
+++ b/ocrweb/rapidocr_onnxruntime/__init__.py
@@ -1,4 +1,4 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
-from .rapid_ocr_api import RapidOCR
\ No newline at end of file
+from .rapid_ocr_api import RapidOCR
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/config.yaml b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/config.yaml
index 79e1a02..0ec669e 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/config.yaml
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/config.yaml
@@ -5,7 +5,6 @@ use_cuda: false
CUDAExecutionProvider:
device_id: 0
arena_extend_strategy: kNextPowerOfTwo
- gpu_mem_limit: 2 * 1024 * 1024 * 1024
cudnn_conv_algo_search: EXHAUSTIVE
do_copy_in_default_stream: true
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/text_cls.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/text_cls.py
index 89b46f2..6ea2e4b 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/text_cls.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/text_cls.py
@@ -19,7 +19,6 @@ from typing import List
import cv2
import numpy as np
-import onnxruntime as ort
try:
from .utils import ClsPostProcess, read_yaml, OrtInferSession
@@ -38,29 +37,6 @@ class TextClassifier(object):
self.session = session_instance.session
self.input_name = session_instance.get_input_name()
- def resize_norm_img(self, img):
- img_c, img_h, img_w = self.cls_image_shape
- h, w = img.shape[:2]
- ratio = w / float(h)
- if math.ceil(img_h * ratio) > img_w:
- resized_w = img_w
- else:
- resized_w = int(math.ceil(img_h * ratio))
-
- resized_image = cv2.resize(img, (resized_w, img_h))
- resized_image = resized_image.astype('float32')
- if img_c == 1:
- resized_image = resized_image / 255
- resized_image = resized_image[np.newaxis, :]
- else:
- resized_image = resized_image.transpose((2, 0, 1)) / 255
-
- resized_image -= 0.5
- resized_image /= 0.5
- padding_im = np.zeros((img_c, img_h, img_w), dtype=np.float32)
- padding_im[:, :, :resized_w] = resized_image
- return padding_im
-
def __call__(self, img_list: List[np.ndarray]):
if isinstance(img_list, np.ndarray):
img_list = [img_list]
@@ -101,6 +77,29 @@ class TextClassifier(object):
img_list[indices[beg_img_no + rno]], 1)
return img_list, cls_res, elapse
+ def resize_norm_img(self, img):
+ img_c, img_h, img_w = self.cls_image_shape
+ h, w = img.shape[:2]
+ ratio = w / float(h)
+ if math.ceil(img_h * ratio) > img_w:
+ resized_w = img_w
+ else:
+ resized_w = int(math.ceil(img_h * ratio))
+
+ resized_image = cv2.resize(img, (resized_w, img_h))
+ resized_image = resized_image.astype('float32')
+ if img_c == 1:
+ resized_image = resized_image / 255
+ resized_image = resized_image[np.newaxis, :]
+ else:
+ resized_image = resized_image.transpose((2, 0, 1)) / 255
+
+ resized_image -= 0.5
+ resized_image /= 0.5
+ padding_im = np.zeros((img_c, img_h, img_w), dtype=np.float32)
+ padding_im[:, :, :resized_w] = resized_image
+ return padding_im
+
if __name__ == "__main__":
parser = argparse.ArgumentParser()
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py
index 570a0b7..b37a8be 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py
@@ -89,4 +89,4 @@ class ClsPostProcess(object):
return decode_out
label = [(self.label_list[idx], 1.0) for idx in label]
- return decode_out, label
+ return decode_out, label
\ No newline at end of file
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/__init__.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/__init__.py
index dd23c8b..502f4a3 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/__init__.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/__init__.py
@@ -1,4 +1,4 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
-from .text_detect import TextDetector
\ No newline at end of file
+from .text_detect import TextDetector
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/config.yaml b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/config.yaml
index c6d4073..9bac869 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/config.yaml
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/config.yaml
@@ -1,11 +1,9 @@
model_path: models/ch_PP-OCRv3_det_infer.onnx
use_cuda: false
-# Details of the params: https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html
CUDAExecutionProvider:
device_id: 0
arena_extend_strategy: kNextPowerOfTwo
- gpu_mem_limit: 2 * 1024 * 1024 * 1024
cudnn_conv_algo_search: EXHAUSTIVE
do_copy_in_default_stream: true
@@ -28,3 +26,4 @@ post_process:
max_candidates: 1000
unclip_ratio: 1.6
use_dilation: true
+ score_mode: "fast"
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/text_detect.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/text_detect.py
index 19b6b31..c3ebf6e 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/text_detect.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/text_detect.py
@@ -20,16 +20,15 @@ import time
import cv2
import numpy as np
-
try:
from .utils import (DBPostProcess, create_operators,
- draw_text_det_res, transform, read_yaml, OrtInferSession)
+ transform, read_yaml, OrtInferSession)
except:
from utils import (DBPostProcess, create_operators,
- draw_text_det_res, transform, read_yaml, OrtInferSession)
+ transform, read_yaml, OrtInferSession)
-class TextDetector(object):
+class TextDetector():
def __init__(self, config):
self.preprocess_op = create_operators(config['pre_process'])
self.postprocess_op = DBPostProcess(**config['post_process'])
@@ -121,6 +120,8 @@ if __name__ == "__main__":
img = cv2.imread(args.image_path)
dt_boxes, elapse = text_detector(img)
+
+ from utils import draw_text_det_res
src_im = draw_text_det_res(dt_boxes, args.image_path)
cv2.imwrite('det_results.jpg', src_im)
print('The det_results.jpg has been saved in the current directory.')
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py
index 0cdae63..d6511b1 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py
@@ -16,19 +16,19 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
-from pathlib import Path
import sys
import warnings
+from pathlib import Path
import cv2
import numpy as np
import pyclipper
import six
import yaml
+from onnxruntime import (GraphOptimizationLevel, InferenceSession,
+ SessionOptions, get_available_providers, get_device)
from shapely.geometry import Polygon
-from onnxruntime import (get_available_providers, get_device,
- SessionOptions, InferenceSession,
- GraphOptimizationLevel)
+
root_dir = Path(__file__).resolve().parent.parent
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/__init__.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/__init__.py
index 686ffdd..ea54cd3 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/__init__.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/__init__.py
@@ -1 +1,4 @@
+# -*- encoding: utf-8 -*-
+# @Author: SWHL
+# @Contact: [email protected]
from .text_recognize import TextRecognizer
\ No newline at end of file
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml
index 1526f9b..4b71491 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/config.yaml
@@ -5,10 +5,8 @@ use_cuda: false
CUDAExecutionProvider:
device_id: 0
arena_extend_strategy: kNextPowerOfTwo
- gpu_mem_limit: 2 * 1024 * 1024 * 1024
cudnn_conv_algo_search: EXHAUSTIVE
do_copy_in_default_stream: true
-rec_img_shape: [3, 32, 320]
+rec_img_shape: [3, 48, 320]
rec_batch_num: 6
-keys_path: resources/rec_dict/ppocr_keys_v1.txt
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/text_recognize.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/text_recognize.py
index 4d306aa..df1bda0 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/text_recognize.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/text_recognize.py
@@ -116,4 +116,4 @@ if __name__ == "__main__":
img = cv2.imread(args.image_path)
rec_res, predict_time = text_recognizer(img)
- print(f'识别结果: {rec_res}\t cost: {predict_time}s')
+ print(f'rec result: {rec_res}\t cost: {predict_time}s')
diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py
index 954c4f8..0860460 100644
--- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py
+++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py
@@ -1,14 +1,14 @@
# -*- encoding: utf-8 -*-
# @Author: SWHL
# @Contact: [email protected]
-from pathlib import Path
import warnings
+from pathlib import Path
import numpy as np
import yaml
-from onnxruntime import (get_available_providers, get_device,
- SessionOptions, InferenceSession,
- GraphOptimizationLevel)
+from onnxruntime import (GraphOptimizationLevel, InferenceSession,
+ SessionOptions, get_available_providers, get_device)
+
root_dir = Path(__file__).resolve().parent.parent
diff --git a/ocrweb/rapidocr_onnxruntime/rapid_ocr_api.py b/ocrweb/rapidocr_onnxruntime/rapid_ocr_api.py
index 9ade4d5..4261223 100644
--- a/ocrweb/rapidocr_onnxruntime/rapid_ocr_api.py
+++ b/ocrweb/rapidocr_onnxruntime/rapid_ocr_api.py
@@ -4,7 +4,6 @@
import copy
import importlib
import sys
-import time
from pathlib import Path
import cv2
@@ -55,26 +54,23 @@ class RapidOCR():
if not self.use_text_det \
or h <= self.min_height \
or use_limit_ratio:
- start_time = time.time()
dt_boxes, img_crop_list = self.get_boxes_img_without_det(img, h, w)
- crop_elapse = time.time() - start_time
- det_elapse = 0
+ det_elapse = 0.0
else:
dt_boxes, det_elapse = self.text_detector(img)
if dt_boxes is None or len(dt_boxes) < 1:
- return None, img, None
+ return None
+
if self.print_verbose:
print(f'dt_boxes num: {len(dt_boxes)}, elapse: {det_elapse}')
dt_boxes = self.sorted_boxes(dt_boxes)
-
- start_time = time.time()
img_crop_list = self.get_crop_img_list(img, dt_boxes)
- crop_elapse = time.time() - start_time
- cls_elapse = 0
+ cls_elapse = 0.0
if self.use_angle_cls:
img_crop_list, _, cls_elapse = self.text_cls(img_crop_list)
+
if self.print_verbose:
print(f'cls num: {len(img_crop_list)}, elapse: {cls_elapse}')
@@ -82,17 +78,11 @@ class RapidOCR():
if self.print_verbose:
print(f'rec_res num: {len(rec_res)}, elapse: {rec_elapse}')
- start_time = time.time()
filter_boxes, filter_rec_res = self.filter_boxes_rec_by_score(dt_boxes,
rec_res)
fina_result = [[dt.tolist(), rec[0], str(rec[1])]
for dt, rec in zip(filter_boxes, filter_rec_res)]
- filter_elapse = time.time() - start_time
- elapse_part = [f'{det_elapse:.4f}',
- f'{(cls_elapse+crop_elapse):.4f}',
- f'{(rec_elapse+filter_elapse):.4f}'
- ]
- return fina_result, img, elapse_part
+ return fina_result, [det_elapse, cls_elapse, rec_elapse]
@staticmethod
def read_yaml(yaml_path):
diff --git a/ocrweb/task.py b/ocrweb/task.py
index 29f7eeb..84201a4 100644
--- a/ocrweb/task.py
+++ b/ocrweb/task.py
@@ -23,7 +23,7 @@ def detect_recognize(image_path, is_api=False):
else:
raise TypeError(f'{image_path} is not str or ndarray.')
- final_result, img, elapse_part = text_sys(image)
+ final_result, elapse_part = text_sys(copy.deepcopy(image))
if is_api:
final_reuslt_json = json.dumps(final_result,
@@ -33,7 +33,7 @@ def detect_recognize(image_path, is_api=False):
if final_result is None:
elapse, elapse_part = 0, ''
- img_str = img_to_base64(img)
+ img_str = img_to_base64(image)
rec_res_data = json.dumps([], indent=2, ensure_ascii=False)
else:
boxes, txts, scores = list(zip(*final_result))
@@ -42,11 +42,11 @@ def detect_recognize(image_path, is_api=False):
indent=2,
ensure_ascii=False)
- det_im = draw_text_det_res(np.array(boxes), img)
+ det_im = draw_text_det_res(np.array(boxes), image)
img_str = img_to_base64(det_im)
elapse = reduce(lambda x, y: float(x)+float(y), elapse_part)
- elapse_part = ','.join([str(x) for x in elapse_part])
+ elapse_part = ','.join([f'{x:.4f}' for x in elapse_part])
return img_str, elapse, elapse_part, rec_res_data
diff --git a/python/demo.py b/python/demo.py
index a742646..e3ca13a 100644
--- a/python/demo.py
+++ b/python/demo.py
@@ -92,7 +92,8 @@ if __name__ == '__main__':
image_path = 'test_images/ch_en_num.jpg'
img = cv2.imread(image_path)
- result = text_sys(img)
+ result, elapse_list = text_sys(img)
print(result)
+ print(elapse_list)
visualize(image_path, result, font_path='resources/fonts/FZYTK.TTF')
diff --git a/python/rapidocr_onnxruntime/rapid_ocr_api.py b/python/rapidocr_onnxruntime/rapid_ocr_api.py
index 0eadb57..4261223 100644
--- a/python/rapidocr_onnxruntime/rapid_ocr_api.py
+++ b/python/rapidocr_onnxruntime/rapid_ocr_api.py
@@ -55,30 +55,34 @@ class RapidOCR():
or h <= self.min_height \
or use_limit_ratio:
dt_boxes, img_crop_list = self.get_boxes_img_without_det(img, h, w)
+ det_elapse = 0.0
else:
- dt_boxes, elapse = self.text_detector(img)
+ dt_boxes, det_elapse = self.text_detector(img)
if dt_boxes is None or len(dt_boxes) < 1:
return None
+
if self.print_verbose:
- print(f'dt_boxes num: {len(dt_boxes)}, elapse: {elapse}')
+ print(f'dt_boxes num: {len(dt_boxes)}, elapse: {det_elapse}')
dt_boxes = self.sorted_boxes(dt_boxes)
img_crop_list = self.get_crop_img_list(img, dt_boxes)
+ cls_elapse = 0.0
if self.use_angle_cls:
- img_crop_list, _, elapse = self.text_cls(img_crop_list)
+ img_crop_list, _, cls_elapse = self.text_cls(img_crop_list)
+
if self.print_verbose:
- print(f'cls num: {len(img_crop_list)}, elapse: {elapse}')
+ print(f'cls num: {len(img_crop_list)}, elapse: {cls_elapse}')
- rec_res, elapse = self.text_recognizer(img_crop_list)
+ rec_res, rec_elapse = self.text_recognizer(img_crop_list)
if self.print_verbose:
- print(f'rec_res num: {len(rec_res)}, elapse: {elapse}')
+ print(f'rec_res num: {len(rec_res)}, elapse: {rec_elapse}')
filter_boxes, filter_rec_res = self.filter_boxes_rec_by_score(dt_boxes,
rec_res)
fina_result = [[dt.tolist(), rec[0], str(rec[1])]
for dt, rec in zip(filter_boxes, filter_rec_res)]
- return fina_result
+ return fina_result, [det_elapse, cls_elapse, rec_elapse]
@staticmethod
def read_yaml(yaml_path):
diff --git a/python/rapidocr_openvino/rapid_ocr_api.py b/python/rapidocr_openvino/rapid_ocr_api.py
index 0566206..4261223 100644
--- a/python/rapidocr_openvino/rapid_ocr_api.py
+++ b/python/rapidocr_openvino/rapid_ocr_api.py
@@ -14,7 +14,7 @@ root_dir = Path(__file__).resolve().parent
sys.path.append(str(root_dir))
-class RapidOCR(object):
+class RapidOCR():
def __init__(self, config_path=str(root_dir / 'config.yaml')):
super(RapidOCR).__init__()
if not Path(config_path).exists():
@@ -55,30 +55,34 @@ class RapidOCR(object):
or h <= self.min_height \
or use_limit_ratio:
dt_boxes, img_crop_list = self.get_boxes_img_without_det(img, h, w)
+ det_elapse = 0.0
else:
- dt_boxes, elapse = self.text_detector(img)
+ dt_boxes, det_elapse = self.text_detector(img)
if dt_boxes is None or len(dt_boxes) < 1:
return None
+
if self.print_verbose:
- print(f'dt_boxes num: {len(dt_boxes)}, elapse: {elapse}')
+ print(f'dt_boxes num: {len(dt_boxes)}, elapse: {det_elapse}')
dt_boxes = self.sorted_boxes(dt_boxes)
img_crop_list = self.get_crop_img_list(img, dt_boxes)
+ cls_elapse = 0.0
if self.use_angle_cls:
- img_crop_list, _, elapse = self.text_cls(img_crop_list)
+ img_crop_list, _, cls_elapse = self.text_cls(img_crop_list)
+
if self.print_verbose:
- print(f'cls num: {len(img_crop_list)}, elapse: {elapse}')
+ print(f'cls num: {len(img_crop_list)}, elapse: {cls_elapse}')
- rec_res, elapse = self.text_recognizer(img_crop_list)
+ rec_res, rec_elapse = self.text_recognizer(img_crop_list)
if self.print_verbose:
- print(f'rec_res num: {len(rec_res)}, elapse: {elapse}')
+ print(f'rec_res num: {len(rec_res)}, elapse: {rec_elapse}')
filter_boxes, filter_rec_res = self.filter_boxes_rec_by_score(dt_boxes,
rec_res)
fina_result = [[dt.tolist(), rec[0], str(rec[1])]
for dt, rec in zip(filter_boxes, filter_rec_res)]
- return fina_result
+ return fina_result, [det_elapse, cls_elapse, rec_elapse]
@staticmethod
def read_yaml(yaml_path):