Skip to content

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

Add the show of the inference time to the web

SWHL75786513b8aa
diff --git a/assets/ocrweb_time.jpg b/assets/ocrweb_time.jpg
new file mode 100644
index 0000000..6a45011
Binary files /dev/null and b/assets/ocrweb_time.jpg differ
diff --git a/ocrweb/README.md b/ocrweb/README.md
index 9db5067..5cf834d 100755
--- a/ocrweb/README.md
+++ b/ocrweb/README.md
@@ -1,3 +1,18 @@
+### 相关问题
+1. 各个阶段使用的模型有哪些?
+ 可以从`task.py`中看到
+ ```python
+ det_model_path = 'resources/models/ch_PP-OCRv2_det_infer.onnx'
+ cls_model_path = 'resources/models/ch_ppocr_mobile_v2.0_cls_infer.onnx'
+ rec_model_path = 'resources/models/ch_ppocr_mobile_v2.0_rec_infer.onnx'
+ ```
+2. 网页上显示的推理时间可以具体解释一下吗?
+ <div align="center">
+ <img src="../assets/ocrweb_time.jpg" width="80%" height="80%">
+ </div>
+3. OCR各个阶段的参数值都从哪里可以看到?
+ - 从各个阶段的目录下对应类的初始化函数可以看到
+
### 运行
1. 安装`requirements.txt`下相关包
```shell
@@ -11,7 +26,7 @@
```shell
python main.py
```
-4. 打开`http://127.0.0.1:9005/`,即可
+4. 打开`http://127.0.0.1:9003/`,即可
### 目录结构
```text
diff --git a/ocrweb/main.py b/ocrweb/main.py
index 3752c09..a7f198c 100644
--- a/ocrweb/main.py
+++ b/ocrweb/main.py
@@ -26,9 +26,12 @@ def ocr():
if request.method == 'POST':
url_get = request.get_json()
url_get = str(url_get).split(',')[1]
+
image = base64.b64decode(url_get)
nparr = np.frombuffer(image, np.uint8)
image = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
+ if image.ndim == 2:
+ image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
return detect_recognize(image)
diff --git a/ocrweb/resources/ch_ppocr_mobile_v2_cls/text_cls.py b/ocrweb/resources/ch_ppocr_mobile_v2_cls/text_cls.py
index e6974ca..c283561 100644
--- a/ocrweb/resources/ch_ppocr_mobile_v2_cls/text_cls.py
+++ b/ocrweb/resources/ch_ppocr_mobile_v2_cls/text_cls.py
@@ -76,6 +76,7 @@ class TextClassifier(object):
return img_list
def __call__(self, image_dir):
+ elapse = 0
if isinstance(image_dir[0], (list, str)):
img_list = self.load_image(image_dir)
else:
@@ -83,16 +84,17 @@ class TextClassifier(object):
img_list = copy.deepcopy(img_list)
img_num = len(img_list)
+
# Calculate the aspect ratio of all text bars
width_list = []
for img in img_list:
width_list.append(img.shape[1] / float(img.shape[0]))
+
# Sorting can speed up the cls process
indices = np.argsort(np.array(width_list))
cls_res = [['', 0.0]] * img_num
batch_num = self.cls_batch_num
- elapse = 0
for beg_img_no in range(0, img_num, batch_num):
end_img_no = min(img_num, beg_img_no + batch_num)
norm_img_batch = []
@@ -112,7 +114,6 @@ class TextClassifier(object):
prob_out = self.session.run(None, onnx_inputs)[0]
cls_result = self.postprocess_op(prob_out)
- elapse += time.time() - starttime
for rno in range(len(cls_result)):
label, score = cls_result[rno]
@@ -120,6 +121,8 @@ class TextClassifier(object):
if '180' in label and score > self.cls_thresh:
img_list[indices[beg_img_no + rno]] = cv2.rotate(
img_list[indices[beg_img_no + rno]], 1)
+
+ elapse += time.time() - starttime
return img_list, cls_res, elapse
diff --git a/ocrweb/resources/ch_ppocr_mobile_v2_det/text_detect.py b/ocrweb/resources/ch_ppocr_mobile_v2_det/text_detect.py
index 32673be..b6f9f51 100644
--- a/ocrweb/resources/ch_ppocr_mobile_v2_det/text_detect.py
+++ b/ocrweb/resources/ch_ppocr_mobile_v2_det/text_detect.py
@@ -94,7 +94,7 @@ class TextDetector(object):
return points
def filter_tag_det_res(self, dt_boxes, image_shape):
- img_height, img_width = image_shape[0:2]
+ img_height, img_width = image_shape[:2]
dt_boxes_new = []
for box in dt_boxes:
box = self.order_points_clockwise(box)
@@ -107,17 +107,11 @@ class TextDetector(object):
dt_boxes = np.array(dt_boxes_new)
return dt_boxes
- def filter_tag_det_res_only_clip(self, dt_boxes, image_shape):
- img_height, img_width = image_shape[0:2]
- dt_boxes_new = []
- for box in dt_boxes:
- box = self.clip_det_res(box, img_height, img_width)
- dt_boxes_new.append(box)
- dt_boxes = np.array(dt_boxes_new)
- return dt_boxes
-
def __call__(self, img):
- ori_im = img.copy()
+ starttime = time.time()
+
+ ori_shape = img.shape[:2]
+
data = {'image': img}
data = transform(data, self.preprocess_op)
img, shape_list = data
@@ -127,13 +121,13 @@ class TextDetector(object):
img = img.astype(np.float32)
shape_list = np.expand_dims(shape_list, axis=0)
- starttime = time.time()
preds = self.session.run([self.output_name],
{self.input_name: img})
post_result = self.postprocess_op(preds[0], shape_list)
dt_boxes = post_result[0]['points']
- dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
+ dt_boxes = self.filter_tag_det_res(dt_boxes, ori_shape)
+
elapse = time.time() - starttime
return dt_boxes, elapse
diff --git a/ocrweb/resources/ch_ppocr_mobile_v2_rec/text_recognize.py b/ocrweb/resources/ch_ppocr_mobile_v2_rec/text_recognize.py
index 54aece6..fa827e9 100644
--- a/ocrweb/resources/ch_ppocr_mobile_v2_rec/text_recognize.py
+++ b/ocrweb/resources/ch_ppocr_mobile_v2_rec/text_recognize.py
@@ -86,21 +86,23 @@ class TextRecognizer(object):
return img_list
def __call__(self, image_dir):
+ elapse = 0
if not isinstance(image_dir, list):
img_list = self.load_image(image_dir)
else:
img_list = image_dir
img_num = len(img_list)
+
# Calculate the aspect ratio of all text bars
width_list = []
for img in img_list:
width_list.append(img.shape[1] / float(img.shape[0]))
+
# Sorting can speed up the recognition process
indices = np.argsort(np.array(width_list))
rec_res = [['', 0.0]] * img_num
batch_num = self.rec_batch_num
- elapse = 0
for beg_img_no in range(0, img_num, batch_num):
end_img_no = min(img_num, beg_img_no + batch_num)
norm_img_batch = []
diff --git a/ocrweb/resources/rapidOCR.py b/ocrweb/resources/rapidOCR.py
index 2b4ef83..cc396f0 100644
--- a/ocrweb/resources/rapidOCR.py
+++ b/ocrweb/resources/rapidOCR.py
@@ -2,6 +2,7 @@
# -*- encoding: utf-8 -*-
import copy
from pathlib import Path
+import time
import cv2
import numpy as np
@@ -46,7 +47,9 @@ class TextSystem(object):
cls_model_path=None) -> None:
super(TextSystem).__init__()
self.text_detector = TextDetector(det_model_path)
+
self.text_recognizer = TextRecognizer(rec_model_path)
+
self.use_angle_cls = use_angle_cls
if self.use_angle_cls:
self.text_classifier = TextClassifier(cls_model_path)
@@ -105,79 +108,39 @@ class TextSystem(object):
_boxes[i + 1] = tmp
return _boxes
- @staticmethod
- def load_image(image_path):
- img, flag = check_and_read_gif(image_path)
- if not flag:
- img = cv2.imread(image_path)
- if img is None:
- raise ValueError(f"error in loading image:{image_path}")
- return img
-
- def __call__(self, image_path):
- # img = self.load_image(image_path)
- img = image_path
- dt_boxes, elapse = self.text_detector(img)
- print("dt_boxes num : {}, elapse : {}".format(
- len(dt_boxes), elapse))
+ def __call__(self, img):
+ dt_boxes, det_elapse = self.text_detector(img)
+ print(f"dt_boxes num: {len(dt_boxes)}, elapse: {det_elapse}")
if dt_boxes is None or dt_boxes.size == 0:
return None, None, img
- img_crop_list = []
+ start_time = time.time()
+ img_crop_list = []
dt_boxes = self.sorted_boxes(dt_boxes)
-
for bno in range(len(dt_boxes)):
tmp_box = copy.deepcopy(dt_boxes[bno])
img_crop = self.get_rotate_crop_image(img, tmp_box)
img_crop_list.append(img_crop)
+ crop_elapse = time.time() - start_time
+
if self.use_angle_cls:
- img_crop_list, angle_list, elapse = self.text_classifier(
- img_crop_list)
- print("cls num : {}, elapse : {}".format(
- len(img_crop_list), elapse))
-
- rec_res, elapse = self.text_recognizer(img_crop_list)
- print("rec_res num : {}, elapse : {}".format(
- len(rec_res), elapse))
- # self.print_draw_crop_rec_res(img_crop_list, rec_res)
+ img_crop_list, _, cls_elapse = self.text_classifier(img_crop_list)
+ print(f"cls num : {len(img_crop_list)}, elapse: {cls_elapse}")
+
+ rec_res, rec_elapse = self.text_recognizer(img_crop_list)
+ print(f"rec_res num: {len(rec_res)}, elapse: {rec_elapse}")
+
+ start_time = time.time()
filter_boxes, filter_rec_res = [], []
for box, rec_reuslt in zip(dt_boxes, rec_res):
text, score = rec_reuslt
if score >= drop_score:
filter_boxes.append(box)
filter_rec_res.append(rec_reuslt)
- return filter_boxes, filter_rec_res, img
-
-
-if __name__ == '__main__':
- # 文本检测+方向分类+文本识别
-
- # v1.0
- # from ch_ppocr_mobile_v1_det import TextDetector
-
- # v2.0 超轻量
- # from ch_ppocr_mobile_v2_cls import TextClassifier
- # from ch_ppocr_mobile_v2_det import TextDetector
- # from ch_ppocr_mobile_v2_rec import TextRecognizer
-
- # det_model_path = 'models/ch_ppocr_mobile_v2_det_train.onnx'
- # cls_model_path = 'models/ch_ppocr_mobile_v2.0_cls_train.onnx'
- # rec_model_path = 'models/ch_ppocr_mobile_v2.0_rec_pre.onnx'
-
- # v2.0 通用模型
- from ch_ppocr_server_v2_det import TextDetector
- from ch_ppocr_server_v2_rec import TextRecognizer
-
- from ch_ppocr_mobile_v2_cls import TextClassifier
-
- det_model_path = 'models/ch_ppocr_server_v2.0_det_train.onnx'
- cls_model_path = 'models/ch_ppocr_mobile_v2.0_cls_train.onnx'
- rec_model_path = 'models/ch_ppocr_server_v2.0_rec_pre.onnx'
-
- image_path = r'test_images/long1.jpg'
+ filter_elapse = time.time() - start_time
- text_sys = TextSystem(det_model_path, rec_model_path,
- use_angle_cls=True,
- cls_model_path=cls_model_path)
- dt_boxes, rec_res = text_sys(image_path)
- visualize(image_path, dt_boxes, rec_res)
+ elapse_part = [f'{det_elapse:.4f}',
+ f'{(cls_elapse+crop_elapse):.4f}',
+ f'{(rec_elapse+filter_elapse):.4f}'
+ ]
+ return filter_boxes, filter_rec_res, img, elapse_part
diff --git a/ocrweb/task.py b/ocrweb/task.py
index 19f9312..4a43a2f 100644
--- a/ocrweb/task.py
+++ b/ocrweb/task.py
@@ -1,19 +1,20 @@
# !/usr/bin/env python
# -*- encoding: utf-8 -*-
-# @File: worker.py
+# @File: task.py
# @Time: 2021/03/07 20:29:32
-# @Author: Max
+# @Author: SWHL
+# @Contact: [email protected]
import base64
+from functools import reduce
import json
import time
import cv2
import numpy as np
-from resources.rapidOCR import TextSystem, draw_text_det_res
+from resources.rapidOCR import TextSystem, draw_text_det_res, check_and_read_gif
-# 实例化模型
-det_model_path = 'resources/models/ch_ppocr_server_v2.0_det_infer.onnx'
+det_model_path = 'resources/models/ch_PP-OCRv2_det_infer.onnx'
cls_model_path = 'resources/models/ch_ppocr_mobile_v2.0_cls_infer.onnx'
rec_model_path = 'resources/models/ch_ppocr_mobile_v2.0_rec_infer.onnx'
@@ -28,10 +29,10 @@ def detect_recognize(image_path):
image = cv2.imread(image_path)
elif isinstance(image_path, np.ndarray):
image = image_path
+ else:
+ raise TypeError(f'{image_path} is not str or ndarray.')
- t1 = time.time()
- dt_boxes, rec_res, img = text_sys(image)
- t2 = time.time()
+ dt_boxes, rec_res, img, elapse_part = text_sys(image)
if dt_boxes is None or rec_res is None:
temp_rec_res = []
@@ -39,6 +40,7 @@ def detect_recognize(image_path):
indent=2,
ensure_ascii=False)
elapse = 0
+ elapse_part = ''
image = cv2.imencode('.jpg', img)[1]
img = str(base64.b64encode(image))[2:-1]
else:
@@ -54,7 +56,9 @@ def detect_recognize(image_path):
image = cv2.imencode('.jpg', det_im)[1]
img = str(base64.b64encode(image))[2:-1]
- elapse = f'{t2-t1:.3f}'
+ elapse = reduce(lambda x, y: float(x)+float(y), elapse_part)
+ elapse_part = ','.join([str(x) for x in elapse_part])
return json.dumps({'image': img,
- 'elapse': elapse,
+ 'total_elapse': f'{elapse:.4f}',
+ 'elapse_part': elapse_part,
'rec_res': rec_res_data})
diff --git a/ocrweb/templates/index.html b/ocrweb/templates/index.html
index c70c8e8..f37d1e7 100644
--- a/ocrweb/templates/index.html
+++ b/ocrweb/templates/index.html
@@ -8,8 +8,9 @@
</head>
<body>
<h1 align="center">
- RapidOCR Demo &nbsp;
- <a href="https://github.com/RapidAI/RapidOCR" style="font-size: 14px;">欢迎Star</a>
+ <b><i>Rapid🗲OCR Demo</i></b>&nbsp;
+ <a href="https://github.com/RapidAI/RapidOCR" style="font-size: 14px;">欢迎Star</a>&thinsp;
+ <a href="https://github.com/RapidAI/RapidOCR/tree/main/ocrweb#readme" style="font-size:14px">了解Demo</a>
</h1>
<div class='area'>
<span class="uplodNote">图片文件类型支持PNG、JPG、JPEG、BMP,图片大小不超过3M。</span>
@@ -26,7 +27,7 @@
<div class="rightarea" id="recogDiv">
<span id="rec_res" class="span_title">文本识别结果</span>
- <br/> <br/>
+ <br/><br/>
<table id="locTable" class="table" cellspacing="0" cellpadding="5"></table>
</div>
</div>
@@ -86,15 +87,27 @@
$("#detect_img").attr('src', 'data:image/jpeg;base64,'+ data['image']);
$('#detect_img').show();
}
- if (data["elapse"]){
- document.getElementById("rec_res").textContent = "文本识别结果(耗时:" + String(data['elapse']) + "s)"
+
+ if (data["total_elapse"]){
+ document.getElementById("rec_res").textContent = "文本识别结果(Total:" + String(data["total_elapse"]) + "s)"
$("#rec_res").show();
}
+
if (data["rec_res"]){
var rec_res = JSON.parse(data["rec_res"]);
- // 绘制整个结果表格
+ console.log(data['elapse_part']);
+ if (data['elapse_part']){
+ var elapse_list = data['elapse_part'].split(',');
+ }else{
+ var elapse_list = [0, 0, 0];
+ }
var data = "";
+ data += "<tr>" +
+ '<th align="right">Det: ' + elapse_list[0] + 's</th>' +
+ '<th align="center">Cls: ' + elapse_list[1] + 's</th>' +
+ '<th align="left">Rec: ' + elapse_list[2] + 's</th>' +
+ "</tr>"
data += "<tr>" +
'<th bgcolor="#C0C0C0">序号</th>' +
'<th bgcolor="#C0C0C0">识别结果</th>' +