Spaces:
Sleeping
Sleeping
rogerxavier
commited on
Commit
•
f3b7d3b
1
Parent(s):
ce94e7c
Upload 2 files
Browse files- app.py +9 -0
- ofa_ocr.py +91 -0
app.py
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from ofa_ocr import ofa_ocr_gr
|
3 |
+
|
4 |
+
if __name__ == "__main__":
|
5 |
+
gr.close_all()
|
6 |
+
ocr_demo = ofa_ocr_gr()
|
7 |
+
ocr_demo.launch(
|
8 |
+
enable_queue=True,
|
9 |
+
)
|
ofa_ocr.py
ADDED
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
os.system('cd ezocr;'
|
3 |
+
'pip install .; cd ..')
|
4 |
+
|
5 |
+
|
6 |
+
import gradio as gr
|
7 |
+
import pandas as pd
|
8 |
+
from PIL import ImageDraw
|
9 |
+
from easyocrlite import ReaderLite
|
10 |
+
from PIL import Image
|
11 |
+
from modelscope.pipelines import pipeline
|
12 |
+
from modelscope.utils.constant import Tasks
|
13 |
+
from modelscope.outputs import OutputKeys
|
14 |
+
|
15 |
+
# step 1. orc detection to find boxes
|
16 |
+
reader = ReaderLite(gpu=True)
|
17 |
+
|
18 |
+
# step 2. recognize ocr result according to ocr detection results
|
19 |
+
ocr_recognize = pipeline(Tasks.ocr_recognition,
|
20 |
+
model='damo/ofa_ocr-recognition_general_base_zh', model_revision='v1.0.0')
|
21 |
+
|
22 |
+
|
23 |
+
def get_images(img: str, reader: ReaderLite, **kwargs):
|
24 |
+
results = reader.process(img, **kwargs)
|
25 |
+
return results
|
26 |
+
|
27 |
+
|
28 |
+
def draw_boxes(image, bounds, color='red', width=4):
|
29 |
+
draw = ImageDraw.Draw(image)
|
30 |
+
for i, bound in enumerate(bounds):
|
31 |
+
p0, p1, p2, p3 = bound
|
32 |
+
draw.text((p0[0]+5, p0[1]+5), str(i+1), fill=color, align='center')
|
33 |
+
draw.line([*p0, *p1, *p2, *p3, *p0], fill=color, width=width)
|
34 |
+
return image
|
35 |
+
|
36 |
+
|
37 |
+
def ofa_ocr_gr():
|
38 |
+
def ocr_api(img):
|
39 |
+
results = get_images(img, reader, max_size=4000, text_confidence=0.7, text_threshold=0.4,
|
40 |
+
link_threshold=0.4, slope_ths=0., add_margin=0.04)
|
41 |
+
box_list, image_list = zip(*results)
|
42 |
+
draw_boxes(img, box_list)
|
43 |
+
|
44 |
+
ocr_result = []
|
45 |
+
for i, (box, image) in enumerate(zip(box_list, image_list)):
|
46 |
+
image = Image.fromarray(image)
|
47 |
+
result = ocr_recognize(image)[OutputKeys.TEXT][0].replace(" ", "")
|
48 |
+
ocr_result.append([str(i + 1), result.replace(' ', '')])
|
49 |
+
|
50 |
+
result = pd.DataFrame(ocr_result, columns=['Box ID', 'Text'])
|
51 |
+
|
52 |
+
return img, result, box_list ##返回box_list 方便微调模型
|
53 |
+
|
54 |
+
examples = [
|
55 |
+
"http://xingchen-data.oss-cn-zhangjiakou.aliyuncs.com/maas/ocr/qiaodaima.png",
|
56 |
+
"http://xingchen-data.oss-cn-zhangjiakou.aliyuncs.com/maas/ocr/shupai.png",
|
57 |
+
"http://xingchen-data.oss-cn-zhangjiakou.aliyuncs.com/maas/ocr/ocr_essay.jpg",
|
58 |
+
"http://xingchen-data.oss-cn-zhangjiakou.aliyuncs.com/maas/ocr/chinese.jpg",
|
59 |
+
"http://xingchen-data.oss-cn-zhangjiakou.aliyuncs.com/maas/ocr/benpao.jpeg",
|
60 |
+
"http://xingchen-data.oss-cn-zhangjiakou.aliyuncs.com/maas/ocr/gaidao.jpeg",
|
61 |
+
]
|
62 |
+
|
63 |
+
title = "<h1 align='center'>基于OFA的OCR识别的应用</h1>"
|
64 |
+
description = '中文OCR体验区,欢迎上传图片,静待检测文字返回~ 相关OCR代码和模型都已在ModelScope开源,支持finetune,欢迎大家在平台上使用!(注:受资源限制,这里只部署了通用OCR模型。)'
|
65 |
+
|
66 |
+
ocr_input_image = gr.components.Image(label='图片', type='pil')
|
67 |
+
|
68 |
+
ocr_output_image = gr.components.Image(label='图片')
|
69 |
+
ocr_output_text = gr.components.Dataframe(label='OCR结果', headers=['ID', '文本'])
|
70 |
+
|
71 |
+
ocr_demo = gr.Interface(
|
72 |
+
fn=ocr_api,
|
73 |
+
inputs=[ocr_input_image],
|
74 |
+
outputs=[ocr_output_image, ocr_output_text],
|
75 |
+
title=title,
|
76 |
+
description=description,
|
77 |
+
allow_flagging='never',
|
78 |
+
examples=examples,
|
79 |
+
examples_per_page=5,
|
80 |
+
cache_examples=True
|
81 |
+
)
|
82 |
+
|
83 |
+
return ocr_demo
|
84 |
+
|
85 |
+
|
86 |
+
if __name__ == "__main__":
|
87 |
+
ocr_demo = ofa_ocr_gr()
|
88 |
+
ocr_demo.launch(
|
89 |
+
share=True,
|
90 |
+
enable_queue=True,
|
91 |
+
)
|