CnOCR-Demo / app.py
breezedeus's picture
use onnx backend for rec models
c67f5ec
raw
history blame
12 kB
# coding: utf-8
# 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
# 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.
# Ref: https://huggingface.co/spaces/hysts/Manga-OCR/blob/main/app.py
import os
import json
import functools
import gradio as gr
import cv2
import numpy as np
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()
MODELS = {}
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))
if len(crops) < 1:
return
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)
def get_ocr_model(det_model_name, rec_model_name, det_more_configs):
global MODELS
config_str = json.dumps(det_more_configs)
if (det_model_name, rec_model_name, config_str) in MODELS:
return MODELS[(det_model_name, rec_model_name, config_str)]
det_model_name, det_model_backend = det_model_name.split('::')
# rec_model_name, rec_model_backend = rec_model_name.split('::')
rec_model_backend = 'onnx'
model = 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,
)
if len(MODELS) > 50:
MODELS = {}
MODELS[(det_model_name, rec_model_name, config_str)] = model
return model
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')
# 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')
return _visualize_ocr(std_out)
def _visualize_ocr(ocr_outs):
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.append([out['score'], out['text']])
return 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)
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)
if is_single_line:
ocr_out = [ocr.ocr_for_single_line(np.array(img))]
else:
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,
)
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}' for m, b in all_models if b == 'onnx']
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>'
)
example_func = functools.partial(
recognize,
# det_model_name='ch_PP-OCRv3_det::onnx',
rotated_bbox=True,
# use_angle_clf=False,
new_size=768,
box_score_thresh=0.3,
min_box_size=10,
)
examples = [
[
'ch_PP-OCRv3_det::onnx',
True,
'number-densenet_lite_136-fc',
False,
'docs/examples/card1-s.jpg',
],
[
'ch_PP-OCRv3_det::onnx',
True,
'number-densenet_lite_136-fc',
False,
'docs/examples/card2-s.jpg',
],
[
'ch_PP-OCRv3_det::onnx',
True,
'number-densenet_lite_136-fc',
False,
'docs/examples/cy1-s.jpg',
],
[
'ch_PP-OCRv3_det::onnx',
False,
'densenet_lite_136-gru',
False,
'docs/examples/huochepiao.jpeg',
],
[
'ch_PP-OCRv3_det::onnx',
False,
'densenet_lite_136-gru',
False,
'docs/examples/1_res.jpg',
],
[
'db_shufflenet_v2::pytorch',
False,
'en_number_mobile_v2.0',
False,
'docs/examples/en_book1.jpeg',
],
[
'db_shufflenet_v2::pytorch',
False,
'densenet_lite_136-gru',
True,
'docs/examples/beauty0.jpg',
],
]
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',
)
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('### 选择待识别图片')
image_file = gr.Image(label='待识别图片', 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],
)
gr.Examples(
label='示例',
examples=examples,
inputs=[
det_model_name,
is_single_line,
rec_model_name,
use_angle_clf,
image_file,
],
outputs=[out_image, naive_warn, out_texts],
fn=example_func,
cache_examples=os.getenv('CACHE_EXAMPLES') == '1',
)
demo.queue(concurrency_count=4)
demo.launch()
if __name__ == '__main__':
main()