transfer from streamlit to gradio

README.md

@@ -1,14 +1,16 @@
 ---
-title: CnOCR
+title: CnOCR Demo
 emoji: 🅞🅒🅡
 colorFrom: indigo
 colorTo: yellow
-sdk: streamlit
+sdk: gradio
+sdk_version: 3.44.4
 app_file: app.py
 pinned: false
+license: apache-2.0
 ---
 
-# CnOCR
+# CnOCR (Cn-OCR)
 
 [**CnOCR**](https://github.com/breezedeus/cnocr)  is an **Optical Character Recognition (OCR)** toolkit for **Python 3**. It supports recognition of common characters in **English and numbers**, **Simplified Chinese**, **Traditional Chinese** (some models), and **vertical text** recognition. It comes with [**20+ well-trained models**](https://cnocr.readthedocs.io/zh/latest/models/) for different application scenarios and can be used directly after installation. Also, CnOCR provides simple training [commands](https://cnocr.readthedocs.io/zh/latest/train/) for users to train their own models. Welcome to join the WeChat contact group.
 

app.py

@@ -1,5 +1,5 @@
 # coding: utf-8
-# Copyright (C) 2021, [Breezedeus](https://github.com/breezedeus).
+# Copyright (C) 2023, [Breezedeus](https://github.com/breezedeus).
 # Licensed to the Apache Software Foundation (ASF) under one
 # or more contributor license agreements.  See the NOTICE file
 # distributed with this work for additional information
@@ -18,12 +18,10 @@
 # under the License.
 
 import os
-from collections import OrderedDict
 
+import gradio as gr
 import cv2
 import numpy as np
-from PIL import Image
-import streamlit as st
 from cnstd.utils import pil_to_numpy, imsave
 
 from cnocr import CnOcr, DET_AVAILABLE_MODELS, REC_AVAILABLE_MODELS
@@ -31,7 +29,6 @@ from cnocr.utils import set_logger, draw_ocr_results, download
 
 
 logger = set_logger()
-st.set_page_config(layout="wide")
 
 
 def plot_for_debugging(rotated_img, one_out, box_score_thresh, crop_ncols, prefix_fp):
@@ -41,6 +38,8 @@ def plot_for_debugging(rotated_img, one_out, box_score_thresh, crop_ncols, prefi
     rotated_img = rotated_img.copy()
     crops = [info['cropped_img'] for info in one_out]
     print('%d boxes are found' % len(crops))
+    if len(crops) < 1:
+        return
     ncols = crop_ncols
     nrows = math.ceil(len(crops) / ncols)
     fig, ax = plt.subplots(nrows=nrows, ncols=ncols)
@@ -64,10 +63,9 @@ def plot_for_debugging(rotated_img, one_out, box_score_thresh, crop_ncols, prefi
     print('boxes results are save to file %s' % result_fp)
 
 
-@st.cache_resource
 def get_ocr_model(det_model_name, rec_model_name, det_more_configs):
-    det_model_name, det_model_backend = det_model_name
-    rec_model_name, rec_model_backend = rec_model_name
+    det_model_name, det_model_backend = det_model_name.split('::')
+    rec_model_name, rec_model_backend = rec_model_name.split('::')
     return CnOcr(
         det_model_name=det_model_name,
         det_model_backend=det_model_backend,
@@ -78,31 +76,33 @@ def get_ocr_model(det_model_name, rec_model_name, det_more_configs):
 
 
 def visualize_naive_result(img, det_model_name, std_out, box_score_thresh):
+    if len(std_out) < 1:
+        # gr.Warning(f'未检测到文本！')
+        return []
     img = pil_to_numpy(img).transpose((1, 2, 0)).astype(np.uint8)
 
-    plot_for_debugging(img, std_out, box_score_thresh, 2, './streamlit-app')
-    st.subheader('Detection Result')
-    if det_model_name == 'default_det':
-        st.warning('⚠️ Warning: "default_det" 检测模型不返回文本框位置！')
-    cols = st.columns([1, 7, 1])
-    cols[1].image('./streamlit-app-result.png')
+    # plot_for_debugging(img, std_out, box_score_thresh, 2, './streamlit-app')
+    # gr.Markdown('## Detection Result')
+    # if det_model_name == 'naive_det':
+    #     gr.Warning('⚠️ Warning: "naive_det" 检测模型不返回文本框位置！')
+    # cols = st.columns([1, 7, 1])
+    # cols[1].image('./streamlit-app-result.png')
+    #
+    # st.subheader('Recognition Result')
+    # cols = st.columns([1, 7, 1])
+    # cols[1].image('./streamlit-app-crops.png')
 
-    st.subheader('Recognition Result')
-    cols = st.columns([1, 7, 1])
-    cols[1].image('./streamlit-app-crops.png')
-
-    _visualize_ocr(std_out)
+    return _visualize_ocr(std_out)
 
 
 def _visualize_ocr(ocr_outs):
-    st.empty()
-    ocr_res = OrderedDict({'文本': []})
-    ocr_res['得分'] = []
+    if len(ocr_outs) < 1:
+        return
+    ocr_res = []
     for out in ocr_outs:
         # cropped_img = out['cropped_img']  # 检测出的文本框
-        ocr_res['得分'].append(out['score'])
-        ocr_res['文本'].append(out['text'])
-    st.table(ocr_res)
+        ocr_res.append([out['score'], out['text']])
+    return ocr_res
 
 
 def visualize_result(img, ocr_outs):
@@ -113,55 +113,27 @@ def visualize_result(img, ocr_outs):
         os.makedirs(os.path.dirname(font_path), exist_ok=True)
         download(url, path=font_path, overwrite=True)
     draw_ocr_results(img, ocr_outs, out_draw_fp, font_path)
-    st.image(out_draw_fp)
-
-
-def main():
-    st.sidebar.header('模型设置')
-    det_models = list(DET_AVAILABLE_MODELS.all_models())
-    det_models.append(('naive_det', 'onnx'))
-    det_models.sort()
-    det_model_name = st.sidebar.selectbox(
-        '选择检测模型', det_models, index=det_models.index(('ch_PP-OCRv3_det', 'onnx'))
-    )
-
-    all_models = list(REC_AVAILABLE_MODELS.all_models())
-    all_models.sort()
-    idx = all_models.index(('densenet_lite_136-fc', 'onnx'))
-    rec_model_name = st.sidebar.selectbox('选择识别模型', all_models, index=idx)
-
-    st.sidebar.subheader('检测参数')
-    rotated_bbox = st.sidebar.checkbox('是否检测带角度文本框', value=True)
-    use_angle_clf = st.sidebar.checkbox('是否使用角度预测模型校正文本框', value=False)
-    new_size = st.sidebar.slider(
-        'resize 后图片（长边）大小', min_value=124, max_value=4096, value=768
-    )
-    box_score_thresh = st.sidebar.slider(
-        '得分阈值（低于阈值的结果会被过滤掉）', min_value=0.05, max_value=0.95, value=0.3
-    )
-    min_box_size = st.sidebar.slider(
-        '框大小阈值（更小的文本框会被过滤掉）', min_value=4, max_value=50, value=10
-    )
-    # std = get_std_model(det_model_name, rotated_bbox, use_angle_clf)
-
-    # st.sidebar.markdown("""---""")
-    # st.sidebar.header('CnOcr 设置')
+    return out_draw_fp
+
+
+def recognize(
+    det_model_name,
+    is_single_line,
+    rec_model_name,
+    rotated_bbox,
+    use_angle_clf,
+    new_size,
+    box_score_thresh,
+    min_box_size,
+    image_file,
+):
+    img = image_file.convert('RGB')
     det_more_configs = dict(rotated_bbox=rotated_bbox, use_angle_clf=use_angle_clf)
     ocr = get_ocr_model(det_model_name, rec_model_name, det_more_configs)
 
-    st.markdown('# 开源Python OCR工具 ' '[CnOCR](https://github.com/breezedeus/cnocr)')
-    st.markdown('> 详细说明参见：[CnOCR 文档](https://cnocr.readthedocs.io/) ；'
-                '欢迎加入 [交流群](https://www.breezedeus.com/join-group) ；'
-                '作者：[breezedeus](https://www.breezedeus.com), [Github](https://github.com/breezedeus) 。')
-    st.markdown('')
-    st.subheader('选择待检测图片')
-    content_file = st.file_uploader('', type=["png", "jpg", "jpeg", "webp"])
-    if content_file is None:
-        st.stop()
-
-    try:
-        img = Image.open(content_file).convert('RGB')
-
+    if is_single_line:
+        ocr_out = [ocr.ocr_for_single_line(np.array(img))]
+    else:
         ocr_out = ocr.ocr(
             img,
             return_cropped_image=True,
@@ -170,13 +142,112 @@ def main():
             box_score_thresh=box_score_thresh,
             min_box_size=min_box_size,
         )
-        if det_model_name[0] == 'naive_det':
-            visualize_naive_result(img, det_model_name[0], ocr_out, box_score_thresh)
-        else:
-            visualize_result(img, ocr_out)
 
-    except Exception as e:
-        st.error(e)
+    det_model_name, det_model_backend = det_model_name.split('::')
+    if is_single_line or det_model_name == 'naive_det':
+        out_texts = visualize_naive_result(
+            img, det_model_name, ocr_out, box_score_thresh
+        )
+        if is_single_line:
+            return [
+                gr.update(visible=False),
+                gr.update(visible=False),
+                gr.update(value=out_texts, visible=True),
+            ]
+        return [
+            gr.update(visible=False),
+            gr.update(visible=True),
+            gr.update(value=out_texts, visible=True),
+        ]
+    else:
+        out_img_path = visualize_result(img, ocr_out)
+        return [
+            gr.update(value=out_img_path, visible=True),
+            gr.update(visible=False),
+            gr.update(visible=False),
+        ]
+
+
+def main():
+    det_models = list(DET_AVAILABLE_MODELS.all_models())
+    det_models.append(('naive_det', 'onnx'))
+    det_models.sort()
+    det_models = [f'{m}::{b}' for m, b in det_models]
+
+    all_models = list(REC_AVAILABLE_MODELS.all_models())
+    all_models.sort()
+    all_models = [f'{m}::{b}' for m, b in all_models]
+
+    title = '开源Python OCR工具：'
+    desc = (
+        '<p style="text-align: center">详细说明参见：<a href="https://github.com/breezedeus/CnOCR" target="_blank">Github</a>；'
+        '<a href="https://cnocr.readthedocs.io" target="_blank">在线文档</a>；'
+        '欢迎加入 <a href="https://www.breezedeus.com/join-group" target="_blank">交流群</a>；'
+        '作者：<a href="https://www.breezedeus.com" target="_blank">Breezedeus</a> ，'
+        '<a href="https://github.com/breezedeus" target="_blank">Github</a> 。</p>'
+    )
+
+    with gr.Blocks() as demo:
+        gr.Markdown(
+            f'<h1 style="text-align: center; margin-bottom: 1rem;">{title} <a href="https://github.com/breezedeus/cnocr" target="_blank">CnOCR</a></h1>'
+        )
+        gr.Markdown(desc)
+        with gr.Row(equal_height=False):
+            with gr.Column(min_width=200, variant='panel', scale=1):
+                gr.Markdown('### 模型设置')
+                det_model_name = gr.Dropdown(
+                    label='选择检测模型', choices=det_models, value='ch_PP-OCRv3_det::onnx',
+                )
+                is_single_line = gr.Checkbox(label='单行文字模式（不使用检测模型）', value=False)
+
+                rec_model_name = gr.Dropdown(
+                    label='选择识别模型',
+                    choices=all_models,
+                    value='densenet_lite_136-fc::onnx',
+                )
+
+                gr.Markdown('### 检测参数')
+                rotated_bbox = gr.Checkbox(label='检测带角度文本框', value=True)
+                use_angle_clf = gr.Checkbox(label='使用角度预测模型校正文本框', value=False)
+                new_size = gr.Slider(
+                    label='resize 后图片（长边）大小', minimum=124, maximum=4096, value=768
+                )
+                box_score_thresh = gr.Slider(
+                    label='得分阈值（低于阈值的结果会被过滤掉）', minimum=0.05, maximum=0.95, value=0.3
+                )
+                min_box_size = gr.Slider(
+                    label='框大小阈值（更小的文本框会被过滤掉）', minimum=4, maximum=50, value=10
+                )
+
+            with gr.Column(scale=3, variant='compact'):
+                gr.Markdown('### 选择待检测图片')
+                image_file = gr.Image(label='', type="pil", image_mode='RGB')
+                sub_btn = gr.Button("Submit", variant="primary")
+                out_image = gr.Image(label='识别结果', interactive=False, visible=False)
+                naive_warn = gr.Markdown(
+                    '**⚠️ Warning**: "naive_det" 检测模型不返回文本框位置！', visible=False
+                )
+                out_texts = gr.Dataframe(
+                    headers=['得分', '文本'], label='识别结果', interactive=False, visible=False
+                )
+                sub_btn.click(
+                    recognize,
+                    inputs=[
+                        det_model_name,
+                        is_single_line,
+                        rec_model_name,
+                        rotated_bbox,
+                        use_angle_clf,
+                        new_size,
+                        box_score_thresh,
+                        min_box_size,
+                        image_file,
+                    ],
+                    outputs=[out_image, naive_warn, out_texts],
+                )
+
+    demo.queue(concurrency_count=4)
+    demo.launch()
 
 
 if __name__ == '__main__':

requirements.txt

@@ -1,155 +1,237 @@
 #
-# This file is autogenerated by pip-compile with python 3.8
-# To update, run:
+# This file is autogenerated by pip-compile with Python 3.9
+# by the following command:
 #
-#    pip-compile --output-file=requirements.txt requirements.in
+#    pip-compile --extra-index-url=https://pypi.tuna.tsinghua.edu.cn/simple --index-url=https://mirrors.aliyun.com/pypi/simple --output-file=requirements.txt requirements.in
 #
---index-url https://pypi.doubanio.com/simple
 --extra-index-url https://pypi.org/simple
 
-absl-py==0.13.0
-    # via tensorboard
-aiohttp==3.7.4.post0
+aiohttp==3.8.4
     # via fsspec
-async-timeout==3.0.1
+aiosignal==1.3.1
     # via aiohttp
-attrs==21.2.0
+albumentations==1.3.1
+    # via -r requirements.in
+appdirs==1.4.4
+    # via wandb
+async-timeout==4.0.2
+    # via aiohttp
+attrs==23.1.0
     # via aiohttp
-cachetools==4.2.2
-    # via google-auth
-certifi==2020.4.5.1
-    # via requests
-chardet==3.0.4
+certifi==2023.5.7
+    # via
+    #   requests
+    #   sentry-sdk
+charset-normalizer==3.1.0
     # via
     #   aiohttp
     #   requests
-click==8.0.1
+click==8.1.3
     # via
     #   -r requirements.in
     #   cnstd
-cnstd==1.2
+    #   wandb
+cnstd>=1.2.3.4
     # via -r requirements.in
+coloredlogs==15.0.1
+    # via onnxruntime
+contourpy==1.1.0
+    # via matplotlib
 cycler==0.11.0
     # via matplotlib
-flatbuffers==2.0
+docker-pycreds==0.4.0
+    # via wandb
+filelock==3.12.2
+    # via
+    #   huggingface-hub
+    #   torch
+flatbuffers==23.5.26
     # via onnxruntime
-fonttools==4.34.4
+fonttools==4.40.0
     # via matplotlib
-fsspec[http]==2021.7.0
-    # via pytorch-lightning
-google-auth==1.35.0
+frozenlist==1.3.3
+    # via
+    #   aiohttp
+    #   aiosignal
+fsspec[http]==2023.6.0
     # via
-    #   google-auth-oauthlib
-    #   tensorboard
-google-auth-oauthlib==0.4.5
-    # via tensorboard
-grpcio==1.39.0
-    # via tensorboard
-idna==2.9
+    #   huggingface-hub
+    #   pytorch-lightning
+gitdb==4.0.10
+    # via gitpython
+gitpython==3.1.34
+    # via wandb
+huggingface-hub==0.15.1
+    # via cnstd
+humanfriendly==10.0
+    # via coloredlogs
+idna==3.4
     # via
     #   requests
     #   yarl
-kiwisolver==1.4.3
+imageio==2.31.3
+    # via scikit-image
+importlib-resources==5.12.0
     # via matplotlib
-markdown==3.3.4
-    # via tensorboard
-matplotlib==3.5.2
-    # via cnstd
-multidict==5.1.0
+jinja2==3.1.2
+    # via torch
+joblib==1.3.2
+    # via scikit-learn
+kiwisolver==1.4.4
+    # via matplotlib
+lazy-loader==0.3
+    # via scikit-image
+lightning-utilities==0.9.0
+    # via pytorch-lightning
+markupsafe==2.1.3
+    # via jinja2
+matplotlib==3.7.1
+    # via
+    #   cnstd
+    #   seaborn
+mpmath==1.3.0
+    # via sympy
+multidict==6.0.4
     # via
     #   aiohttp
     #   yarl
-numpy==1.22.3
+networkx==3.1
+    # via
+    #   scikit-image
+    #   torch
+numpy==1.25.0
     # via
     #   -r requirements.in
+    #   albumentations
     #   cnstd
+    #   contourpy
+    #   imageio
     #   matplotlib
     #   onnx
     #   onnxruntime
     #   opencv-python
+    #   opencv-python-headless
+    #   pandas
     #   pytorch-lightning
+    #   pywavelets
+    #   qudida
+    #   scikit-image
+    #   scikit-learn
     #   scipy
-    #   tensorboard
+    #   seaborn
+    #   shapely
+    #   tifffile
     #   torchmetrics
     #   torchvision
-oauthlib==3.1.1
-    # via requests-oauthlib
-onnx==1.13.0
+onnx==1.14.0
     # via
     #   -r requirements.in
     #   cnstd
-onnxruntime==1.14.0
+onnxruntime==1.15.1
     # via
     #   -r requirements.in
     #   cnstd
-opencv-python==4.6.0.66
+opencv-python==4.7.0.72
     # via cnstd
-packaging==21.0
+opencv-python-headless==4.8.0.76
+    # via
+    #   albumentations
+    #   qudida
+packaging==23.1
     # via
+    #   huggingface-hub
+    #   lightning-utilities
     #   matplotlib
+    #   onnxruntime
     #   pytorch-lightning
+    #   scikit-image
     #   torchmetrics
-pillow==9.1.1
+pandas==2.0.3
+    # via
+    #   cnstd
+    #   seaborn
+pathtools==0.1.2
+    # via wandb
+pillow==9.5.0
     # via
     #   -r requirements.in
     #   cnstd
+    #   imageio
     #   matplotlib
+    #   scikit-image
     #   torchvision
 polygon3==3.0.9.1
     # via cnstd
-protobuf
+protobuf==4.23.3
     # via
     #   onnx
     #   onnxruntime
-    #   tensorboard
-pyasn1==0.4.8
-    # via
-    #   pyasn1-modules
-    #   rsa
-pyasn1-modules==0.2.8
-    # via google-auth
-pyclipper==1.3.0.post3
+    #   wandb
+psutil==5.9.5
+    # via wandb
+pyclipper==1.3.0.post4
     # via cnstd
-pydeprecate==0.3.1
-    # via pytorch-lightning
-pyparsing==2.4.7
+pyparsing==3.1.0
+    # via matplotlib
+python-dateutil==2.8.2
     # via
     #   matplotlib
-    #   packaging
-python-dateutil==2.8.2
-    # via matplotlib
-pytorch-lightning==1.6.3
+    #   pandas
+pytorch-lightning==2.0.8
     # via
     #   -r requirements.in
     #   cnstd
-pyyaml==5.4.1
-    # via pytorch-lightning
-requests==2.23.0
+pytz==2023.3
+    # via pandas
+pywavelets==1.4.1
+    # via scikit-image
+pyyaml==6.0
+    # via
+    #   albumentations
+    #   cnstd
+    #   huggingface-hub
+    #   pytorch-lightning
+    #   wandb
+qudida==0.0.4
+    # via albumentations
+requests==2.31.0
     # via
     #   fsspec
-    #   requests-oauthlib
-    #   tensorboard
-requests-oauthlib==1.3.0
-    # via google-auth-oauthlib
-rsa==4.7.2
-    # via google-auth
-scipy==1.8.1
+    #   huggingface-hub
+    #   torchvision
+    #   wandb
+scikit-image==0.21.0
+    # via albumentations
+scikit-learn==1.3.0
+    # via qudida
+scipy==1.11.1
+    # via
+    #   albumentations
+    #   cnstd
+    #   scikit-image
+    #   scikit-learn
+seaborn==0.12.2
     # via cnstd
-shapely==1.8.2
+sentry-sdk==1.30.0
+    # via wandb
+setproctitle==1.3.2
+    # via wandb
+shapely==2.0.1
     # via cnstd
-six==1.14.0
+six==1.16.0
     # via
-    #   absl-py
-    #   google-auth
-    #   grpcio
-    #   protobuf
+    #   docker-pycreds
     #   python-dateutil
-tensorboard==2.6.0
-    # via pytorch-lightning
-tensorboard-data-server==0.6.1
-    # via tensorboard
-tensorboard-plugin-wit==1.8.0
-    # via tensorboard
+smmap==5.0.0
+    # via gitdb
+sympy==1.12
+    # via
+    #   onnxruntime
+    #   torch
+threadpoolctl==3.2.0
+    # via scikit-learn
+tifffile==2023.8.30
+    # via scikit-image
 torch==2.0.1
     # via
     #   -r requirements.in
@@ -157,37 +239,45 @@ torch==2.0.1
     #   pytorch-lightning
     #   torchmetrics
     #   torchvision
-torchmetrics==0.11.1
-    # via pytorch-lightning
+torchmetrics==0.11.4
+    # via
+    #   -r requirements.in
+    #   pytorch-lightning
 torchvision==0.15.2
     # via
     #   -r requirements.in
     #   cnstd
-tqdm==4.64.0
+tqdm==4.65.0
     # via
     #   -r requirements.in
     #   cnstd
+    #   huggingface-hub
     #   pytorch-lightning
-typing-extensions==4.2.0
+typing-extensions==4.7.0
     # via
-    #   aiohttp
+    #   huggingface-hub
+    #   lightning-utilities
     #   onnx
     #   pytorch-lightning
+    #   qudida
     #   torch
-unidecode==1.3.4
+    #   wandb
+tzdata==2023.3
+    # via pandas
+unidecode==1.3.6
     # via cnstd
-urllib3==1.25.9
-    # via requests
-werkzeug==2.0.1
-    # via tensorboard
-wheel==0.37.0
-    # via tensorboard
-yarl==1.6.3
+urllib3==2.0.3
+    # via
+    #   requests
+    #   sentry-sdk
+wandb==0.15.10
+    # via -r requirements.in
+yarl==1.9.2
     # via aiohttp
+zipp==3.15.0
+    # via importlib-resources
+
+cnocr==2.2.4
 
 # The following packages are considered to be unsafe in a requirements file:
 # setuptools
-
-# for streamlit.io demo
-cnocr==2.2.2.3
-streamlit

streamlit_app.py

@@ -0,0 +1,183 @@
+# coding: utf-8
+# Copyright (C) 2021, [Breezedeus](https://github.com/breezedeus).
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+import os
+from collections import OrderedDict
+
+import cv2
+import numpy as np
+from PIL import Image
+import streamlit as st
+from cnstd.utils import pil_to_numpy, imsave
+
+from cnocr import CnOcr, DET_AVAILABLE_MODELS, REC_AVAILABLE_MODELS
+from cnocr.utils import set_logger, draw_ocr_results, download
+
+
+logger = set_logger()
+st.set_page_config(layout="wide")
+
+
+def plot_for_debugging(rotated_img, one_out, box_score_thresh, crop_ncols, prefix_fp):
+    import matplotlib.pyplot as plt
+    import math
+
+    rotated_img = rotated_img.copy()
+    crops = [info['cropped_img'] for info in one_out]
+    print('%d boxes are found' % len(crops))
+    ncols = crop_ncols
+    nrows = math.ceil(len(crops) / ncols)
+    fig, ax = plt.subplots(nrows=nrows, ncols=ncols)
+    for i, axi in enumerate(ax.flat):
+        if i >= len(crops):
+            break
+        axi.imshow(crops[i])
+    crop_fp = '%s-crops.png' % prefix_fp
+    plt.savefig(crop_fp)
+    print('cropped results are save to file %s' % crop_fp)
+
+    for info in one_out:
+        box, score = info.get('position'), info['score']
+        if score < box_score_thresh:  # score < 0.5
+            continue
+        if box is not None:
+            box = box.astype(int).reshape(-1, 2)
+            cv2.polylines(rotated_img, [box], True, color=(255, 0, 0), thickness=2)
+    result_fp = '%s-result.png' % prefix_fp
+    imsave(rotated_img, result_fp, normalized=False)
+    print('boxes results are save to file %s' % result_fp)
+
+
+@st.cache_resource
+def get_ocr_model(det_model_name, rec_model_name, det_more_configs):
+    det_model_name, det_model_backend = det_model_name
+    rec_model_name, rec_model_backend = rec_model_name
+    return CnOcr(
+        det_model_name=det_model_name,
+        det_model_backend=det_model_backend,
+        rec_model_name=rec_model_name,
+        rec_model_backend=rec_model_backend,
+        det_more_configs=det_more_configs,
+    )
+
+
+def visualize_naive_result(img, det_model_name, std_out, box_score_thresh):
+    img = pil_to_numpy(img).transpose((1, 2, 0)).astype(np.uint8)
+
+    plot_for_debugging(img, std_out, box_score_thresh, 2, './streamlit-app')
+    st.subheader('Detection Result')
+    if det_model_name == 'default_det':
+        st.warning('⚠️ Warning: "default_det" 检测模型不返回文本框位置！')
+    cols = st.columns([1, 7, 1])
+    cols[1].image('./streamlit-app-result.png')
+
+    st.subheader('Recognition Result')
+    cols = st.columns([1, 7, 1])
+    cols[1].image('./streamlit-app-crops.png')
+
+    _visualize_ocr(std_out)
+
+
+def _visualize_ocr(ocr_outs):
+    st.empty()
+    ocr_res = OrderedDict({'文本': []})
+    ocr_res['得分'] = []
+    for out in ocr_outs:
+        # cropped_img = out['cropped_img']  # 检测出的文本框
+        ocr_res['得分'].append(out['score'])
+        ocr_res['文本'].append(out['text'])
+    st.table(ocr_res)
+
+
+def visualize_result(img, ocr_outs):
+    out_draw_fp = './streamlit-app-det-result.png'
+    font_path = 'docs/fonts/simfang.ttf'
+    if not os.path.exists(font_path):
+        url = 'https://huggingface.co/datasets/breezedeus/cnocr-wx-qr-code/resolve/main/fonts/simfang.ttf'
+        os.makedirs(os.path.dirname(font_path), exist_ok=True)
+        download(url, path=font_path, overwrite=True)
+    draw_ocr_results(img, ocr_outs, out_draw_fp, font_path)
+    st.image(out_draw_fp)
+
+
+def main():
+    st.sidebar.header('模型设置')
+    det_models = list(DET_AVAILABLE_MODELS.all_models())
+    det_models.append(('naive_det', 'onnx'))
+    det_models.sort()
+    det_model_name = st.sidebar.selectbox(
+        '选择检测模型', det_models, index=det_models.index(('ch_PP-OCRv3_det', 'onnx'))
+    )
+
+    all_models = list(REC_AVAILABLE_MODELS.all_models())
+    all_models.sort()
+    idx = all_models.index(('densenet_lite_136-fc', 'onnx'))
+    rec_model_name = st.sidebar.selectbox('选择识别模型', all_models, index=idx)
+
+    st.sidebar.subheader('检测参数')
+    rotated_bbox = st.sidebar.checkbox('是否检测带角度文本框', value=True)
+    use_angle_clf = st.sidebar.checkbox('是否使用角度预测模型校正文本框', value=False)
+    new_size = st.sidebar.slider(
+        'resize 后图片（长边）大小', min_value=124, max_value=4096, value=768
+    )
+    box_score_thresh = st.sidebar.slider(
+        '得分阈值（低于阈值的结果会被过滤掉）', min_value=0.05, max_value=0.95, value=0.3
+    )
+    min_box_size = st.sidebar.slider(
+        '框大小阈值（更小的文本框会被过滤掉）', min_value=4, max_value=50, value=10
+    )
+    # std = get_std_model(det_model_name, rotated_bbox, use_angle_clf)
+
+    # st.sidebar.markdown("""---""")
+    # st.sidebar.header('CnOcr 设置')
+    det_more_configs = dict(rotated_bbox=rotated_bbox, use_angle_clf=use_angle_clf)
+    ocr = get_ocr_model(det_model_name, rec_model_name, det_more_configs)
+
+    st.markdown('# 开源Python OCR工具 ' '[CnOCR](https://github.com/breezedeus/cnocr)')
+    st.markdown('> 详细说明参见：[CnOCR 文档](https://cnocr.readthedocs.io/) ；'
+                '欢迎加入 [交流群](https://www.breezedeus.com/join-group) ；'
+                '作者：[breezedeus](https://www.breezedeus.com), [Github](https://github.com/breezedeus) 。')
+    st.markdown('')
+    st.subheader('选择待检测图片')
+    content_file = st.file_uploader('', type=["png", "jpg", "jpeg", "webp"])
+    if content_file is None:
+        st.stop()
+
+    try:
+        img = Image.open(content_file).convert('RGB')
+
+        ocr_out = ocr.ocr(
+            img,
+            return_cropped_image=True,
+            resized_shape=new_size,
+            preserve_aspect_ratio=True,
+            box_score_thresh=box_score_thresh,
+            min_box_size=min_box_size,
+        )
+        if det_model_name[0] == 'naive_det':
+            visualize_naive_result(img, det_model_name[0], ocr_out, box_score_thresh)
+        else:
+            visualize_result(img, ocr_out)
+
+    except Exception as e:
+        st.error(e)
+
+
+if __name__ == '__main__':
+    main()