File size: 11,410 Bytes
cef4f97
 
22c7b5b
cef4f97
 
 
d1d0907
1a87a19
02ff46f
 
 
4967985
7dcbad8
 
 
4326b14
9cf89ef
63a03ad
4326b14
02ff46f
ed456c1
22c7b5b
3b378ec
 
cef4f97
02ff46f
3b378ec
 
00ee90b
cef4f97
4326b14
 
 
 
 
 
ee4b3d0
fa528ee
 
 
 
 
 
4326b14
131353d
4326b14
 
997a61b
7dcbad8
4326b14
 
7dcbad8
4326b14
3b378ec
fa528ee
 
 
9cf89ef
464330e
 
fa528ee
 
 
 
 
9cf89ef
fa528ee
 
 
 
 
4326b14
131353d
7dcbad8
4326b14
 
131353d
4326b14
131353d
4326b14
131353d
4326b14
131353d
4326b14
131353d
4326b14
 
 
 
 
 
 
 
 
 
 
 
9cf89ef
fa528ee
 
 
 
 
 
cef4f97
ee4b3d0
11f2cf1
 
 
 
 
 
 
 
 
cef4f97
 
 
 
 
11f2cf1
 
 
 
cef4f97
 
 
 
 
 
11f2cf1
 
 
 
cef4f97
f35ea73
3497964
d1a7b5b
 
3497964
eeaf186
d1a7b5b
 
3497964
d1a7b5b
 
 
 
 
3497964
d1a7b5b
 
3497964
d1a7b5b
eeaf186
3497964
d1a7b5b
eeaf186
 
d1a7b5b
 
3497964
eeaf186
d1a7b5b
 
eeaf186
d1a7b5b
3497964
d1a7b5b
eeaf186
d1a7b5b
 
 
3f89105
3497964
a197908
131353d
 
ee4b3d0
997a61b
4326b14
fa528ee
 
3df4587
 
ee4b3d0
02ff46f
4326b14
 
 
 
63a03ad
 
7dcbad8
 
 
4326b14
 
 
 
cef4f97
2cd1f0b
4c630fe
 
ff7a2a0
 
448bc9b
 
ff7a2a0
 
 
 
4967985
de261b5
448bc9b
dfc79d4
 
 
 
3405c5a
dfc79d4
 
3405c5a
4c630fe
 
 
 
fa528ee
4c630fe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
131353d
a197908
131353d
3b378ec
4c630fe
fa528ee
ee4b3d0
4c630fe
 
3df4587
4c630fe
02ff46f
cef4f97
 
 
fa528ee
 
 
 
 
 
 
cef4f97
 
 
 
131353d
ee4b3d0
cef4f97
fa528ee
 
 
 
 
 
f34dca6
7dcbad8
e4e4366
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
import gradio as gr
import torch
from transformers import AutoModel, AutoTokenizer, AutoConfig
import os
import base64
import spaces
import io
from PIL import Image
import numpy as np
import yaml
from pathlib import Path
from globe import title, description, modelinfor, joinus, howto
import uuid
import tempfile
import time
import shutil
import cv2
import re


model_name = 'ucaslcl/GOT-OCR2_0'

device = 'cuda' if torch.cuda.is_available() else 'cpu'

tokenizer = AutoTokenizer.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True)
config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
model = AutoModel.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True, low_cpu_mem_usage=True,  device_map=device, use_safetensors=True, pad_token_id=tokenizer.eos_token_id)
model = model.eval().to(device)
model.config.pad_token_id = tokenizer.eos_token_id

UPLOAD_FOLDER = "./uploads"
RESULTS_FOLDER = "./results"

for folder in [UPLOAD_FOLDER, RESULTS_FOLDER]:
    if not os.path.exists(folder):
        os.makedirs(folder)

def image_to_base64(image):
    buffered = io.BytesIO()
    image.save(buffered, format="PNG")
    return base64.b64encode(buffered.getvalue()).decode()


@spaces.GPU()
def process_image(image, task, ocr_type=None, ocr_box=None, ocr_color=None):
    if image is None:
        return "Error: No image provided", None, None
    
    unique_id = str(uuid.uuid4())
    image_path = os.path.join(UPLOAD_FOLDER, f"{unique_id}.png")
    result_path = os.path.join(RESULTS_FOLDER, f"{unique_id}.html")
    
    try:
        if isinstance(image, dict):
            composite_image = image.get("composite")
            if composite_image is not None:
                if isinstance(composite_image, np.ndarray):
                    cv2.imwrite(image_path, cv2.cvtColor(composite_image, cv2.COLOR_RGB2BGR))
                elif isinstance(composite_image, Image.Image):
                    composite_image.save(image_path)
                else:
                    return "Error: Unsupported image format from ImageEditor", None, None
            else:
                return "Error: No composite image found in ImageEditor output", None, None
        elif isinstance(image, np.ndarray):
            cv2.imwrite(image_path, cv2.cvtColor(image, cv2.COLOR_RGB2BGR))
        elif isinstance(image, str):
            shutil.copy(image, image_path)
        else:
            return "Error: Unsupported image format", None, None

        if task == "Plain Text OCR":
            res = model.chat(tokenizer, image_path, ocr_type='ocr')
            return res, None, unique_id
        else:
            if task == "Format Text OCR":
                res = model.chat(tokenizer, image_path, ocr_type='format', render=True, save_render_file=result_path)
            elif task == "Fine-grained OCR (Box)":
                res = model.chat(tokenizer, image_path, ocr_type=ocr_type, ocr_box=ocr_box, render=True, save_render_file=result_path)
            elif task == "Fine-grained OCR (Color)":
                res = model.chat(tokenizer, image_path, ocr_type=ocr_type, ocr_color=ocr_color, render=True, save_render_file=result_path)
            elif task == "Multi-crop OCR":
                res = model.chat_crop(tokenizer, image_path, ocr_type='format', render=True, save_render_file=result_path)
            elif task == "Render Formatted OCR":
                res = model.chat(tokenizer, image_path, ocr_type='format', render=True, save_render_file=result_path)
            
            if os.path.exists(result_path):
                with open(result_path, 'r') as f:
                    html_content = f.read()
                return res, html_content, unique_id
            else:
                return res, None, unique_id
    except Exception as e:
        return f"Error: {str(e)}", None, None
    finally:
        if os.path.exists(image_path):
            os.remove(image_path)
            
def update_image_input(task):
    if task == "Fine-grained OCR (Color)":
        return gr.update(visible=False), gr.update(visible=True), gr.update(visible=True)
    else:
        return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False)

def update_inputs(task):
    if task in ["Plain Text OCR", "Format Text OCR", "Multi-crop OCR", "Render Formatted OCR"]:
        return [
            gr.update(visible=False),
            gr.update(visible=False),
            gr.update(visible=False),
            gr.update(visible=True),
            gr.update(visible=False),
            gr.update(visible=True),
            gr.update(visible=False)
        ]
    elif task == "Fine-grained OCR (Box)":
        return [
            gr.update(visible=True, choices=["ocr", "format"]),
            gr.update(visible=True),
            gr.update(visible=False),
            gr.update(visible=True),
            gr.update(visible=False),
            gr.update(visible=True),
            gr.update(visible=False)
        ]
    elif task == "Fine-grained OCR (Color)":
        return [
            gr.update(visible=True, choices=["ocr", "format"]),
            gr.update(visible=False),
            gr.update(visible=True, choices=["red", "green", "blue"]),
            gr.update(visible=False),
            gr.update(visible=True),
            gr.update(visible=False),
            gr.update(visible=True)
        ]
    
def parse_latex_output(res):
    # Split the input, preserving newlines and empty lines
    lines = re.split(r'(\$\$.*?\$\$)', res, flags=re.DOTALL)
    parsed_lines = []
    in_latex = False
    latex_buffer = []

    for line in lines:
        if line == '\n':
            if in_latex:
                latex_buffer.append(line)
            else:
                parsed_lines.append(line)
            continue

        line = line.strip()
        
        latex_patterns = [r'\{', r'\}', r'\[', r'\]', r'\\', r'\$', r'_', r'^', r'"']
        contains_latex = any(re.search(pattern, line) for pattern in latex_patterns)
        
        if contains_latex:
            if not in_latex:
                in_latex = True
                latex_buffer = ['$$']
            latex_buffer.append(line)
        else:
            if in_latex:
                latex_buffer.append('$$')
                parsed_lines.extend(latex_buffer)
                in_latex = False
                latex_buffer = []
            parsed_lines.append(line)

    if in_latex:
        latex_buffer.append('$$')
        parsed_lines.extend(latex_buffer)

    return '$$\\$$\n'.join(parsed_lines)
                         

def ocr_demo(image, task, ocr_type, ocr_box, ocr_color):
    res, html_content, unique_id = process_image(image, task, ocr_type, ocr_box, ocr_color)
    
    if isinstance(res, str) and res.startswith("Error:"):
        return res, None

    res = res.replace("\\title", "\\title ")
    formatted_res = res
    # formatted_res = parse_latex_output(res)
    
    if html_content:
        encoded_html = base64.b64encode(html_content.encode('utf-8')).decode('utf-8')
        iframe_src = f"data:text/html;base64,{encoded_html}"
        iframe = f'<iframe src="{iframe_src}" width="100%" height="600px"></iframe>'
        download_link = f'<a href="data:text/html;base64,{encoded_html}" download="result_{unique_id}.html">Download Full Result</a>'
        return formatted_res, f"{download_link}<br>{iframe}"
    return formatted_res, None

def cleanup_old_files():
    current_time = time.time()
    for folder in [UPLOAD_FOLDER, RESULTS_FOLDER]:
        for file_path in Path(folder).glob('*'):
            if current_time - file_path.stat().st_mtime > 3600:  # 1 hour
                file_path.unlink()

with gr.Blocks(theme=gr.themes.Base()) as demo:
    with gr.Row():
        gr.Markdown(title)
    with gr.Row():
        with gr.Column(scale=1):
            with gr.Group():                    
                gr.Markdown(description)
        with gr.Column(scale=1):
            with gr.Group():
                gr.Markdown(modelinfor)
                gr.Markdown(joinus)
    with gr.Row():
        with gr.Accordion("How to use Fine-grained OCR (Color)", open=False):
            with gr.Row():
                gr.Image("res/image/howto_1.png", label="Select the Following Parameters")
                gr.Image("res/image/howto_2.png", label="Click on Paintbrush in the Image Editor")
                gr.Image("res/image/howto_3.png", label="Select your Brush Color (Red)")
                gr.Image("res/image/howto_4.png", label="Make a Box Around The Text")
            with gr.Row():
                with gr.Group():
                    gr.Markdown(howto)

    with gr.Row():
        with gr.Column(scale=1):
            with gr.Group():
                image_input = gr.Image(type="filepath", label="Input Image")
                image_editor = gr.ImageEditor(label="Image Editor", type="pil", visible=False)
                task_dropdown = gr.Dropdown(
                    choices=[
                        "Plain Text OCR",
                        "Format Text OCR",
                        "Fine-grained OCR (Box)",
                        "Fine-grained OCR (Color)",
                        "Multi-crop OCR",
                        "Render Formatted OCR"
                    ],
                    label="Select Task",
                    value="Plain Text OCR"
                )
                ocr_type_dropdown = gr.Dropdown(
                    choices=["ocr", "format"],
                    label="OCR Type",
                    visible=False
                )
                ocr_box_input = gr.Textbox(
                    label="OCR Box (x1,y1,x2,y2)",
                    placeholder="[100,100,200,200]",
                    visible=False
                )
                ocr_color_dropdown = gr.Dropdown(
                    choices=["red", "green", "blue"],
                    label="OCR Color",
                    visible=False
                )
                # with gr.Row():
                    # max_new_tokens_slider = gr.Slider(50, 500, step=10, value=150, label="Max New Tokens")
                    # no_repeat_ngram_size_slider = gr.Slider(1, 10, step=1, value=2, label="No Repeat N-gram Size")

                submit_button = gr.Button("Process")
                editor_submit_button = gr.Button("Process Edited Image", visible=False)

        with gr.Column(scale=1):
            with gr.Group():
                output_markdown = gr.Textbox(label="🫴🏻📸GOT-OCR")
                output_html = gr.HTML(label="🫴🏻📸GOT-OCR")

    task_dropdown.change(
        update_inputs,
        inputs=[task_dropdown],
        outputs=[ocr_type_dropdown, ocr_box_input, ocr_color_dropdown, image_input, image_editor, submit_button, editor_submit_button]
    )
    
    task_dropdown.change(
        update_image_input,
        inputs=[task_dropdown],
        outputs=[image_input, image_editor, editor_submit_button]
    )
    
    submit_button.click(
        ocr_demo,
        inputs=[image_input, task_dropdown, ocr_type_dropdown, ocr_box_input, ocr_color_dropdown],
        outputs=[output_markdown, output_html]
    )
    editor_submit_button.click(
        ocr_demo,
        inputs=[image_editor, task_dropdown, ocr_type_dropdown, ocr_box_input, ocr_color_dropdown],
        outputs=[output_markdown, output_html]
    )

if __name__ == "__main__":
    cleanup_old_files()
    demo.launch(ssr_mode = False)