File size: 9,423 Bytes
8959aae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
DeepSeek OCR Service Module
Handles OCR text extraction using DeepSeek-OCR model
"""

import os
import torch
from PIL import Image
from transformers import AutoModelForCausalLM, AutoTokenizer
from typing import Optional, Dict, Any
import logging
from pathlib import Path
from dotenv import load_dotenv

# Load environment variables
load_dotenv()

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

class DeepSeekOCRService:
    """
    Service class for DeepSeek OCR text extraction
    """
    
    def __init__(self, model_name: str = None):
        """
        Initialize the DeepSeek OCR service
        
        Args:
            model_name (str): Hugging Face model name for DeepSeek OCR
        """
        self.model_name = model_name or os.getenv('DEEPSEEK_OCR_MODEL', 'deepseek-ai/DeepSeek-OCR')
        self.model = None
        self.tokenizer = None
        
        # Device configuration - optimized for CPU
        device_config = os.getenv('DEEPSEEK_OCR_DEVICE', 'cpu')
        if device_config == 'auto':
            self.device = "cuda" if torch.cuda.is_available() else "cpu"
        else:
            self.device = device_config
            
        logger.info(f"Initializing DeepSeek OCR on device: {self.device}")
        
    def load_model(self):
        """
        Load the DeepSeek OCR model and tokenizer
        """
        try:
            logger.info(f"Loading DeepSeek OCR model: {self.model_name}")
            self.tokenizer = AutoTokenizer.from_pretrained(
                self.model_name, 
                trust_remote_code=True
            )
            # CPU-optimized model loading
            if self.device == "cpu":
                self.model = AutoModelForCausalLM.from_pretrained(
                    self.model_name, 
                    trust_remote_code=True,
                    torch_dtype=torch.float32,  # Use float32 for CPU
                    low_cpu_mem_usage=True,     # Reduce memory usage
                    device_map="cpu"            # Force CPU usage
                )
            else:
                self.model = AutoModelForCausalLM.from_pretrained(
                    self.model_name, 
                    trust_remote_code=True,
                    torch_dtype=torch.float16 if self.device == "cuda" else torch.float32
                )
            self.model.to(self.device)
            logger.info("DeepSeek OCR model loaded successfully")
        except Exception as e:
            logger.error(f"Failed to load DeepSeek OCR model: {str(e)}")
            raise e
    
    def extract_text_from_image(self, image_path: str, prompt: str = None) -> Dict[str, Any]:
        """
        Extract text from an image using DeepSeek OCR
        
        Args:
            image_path (str): Path to the image file
            prompt (str, optional): Custom prompt for OCR processing
            
        Returns:
            Dict containing extracted text and metadata
        """
        if self.model is None or self.tokenizer is None:
            self.load_model()
        
        try:
            # Load and preprocess the image
            image = Image.open(image_path)
            if image.mode != 'RGB':
                image = image.convert('RGB')
            
            # Use default prompt if none provided
            if prompt is None:
                prompt = "<|grounding|>Extract all text from this image."
            
            # Prepare inputs
            inputs = self.tokenizer(
                prompt, 
                image, 
                return_tensors="pt"
            ).to(self.device)
            
            # Get configuration from environment - CPU optimized defaults
            max_tokens = int(os.getenv('DEEPSEEK_OCR_MAX_TOKENS', '256'))  # Reduced for CPU
            temperature = float(os.getenv('DEEPSEEK_OCR_TEMPERATURE', '0.1'))
            
            # Generate text extraction
            with torch.no_grad():
                outputs = self.model.generate(
                    **inputs,
                    max_new_tokens=max_tokens,
                    do_sample=False,
                    temperature=temperature,
                    pad_token_id=self.tokenizer.eos_token_id
                )
            
            # Decode the output
            extracted_text = self.tokenizer.decode(
                outputs[0], 
                skip_special_tokens=True
            )
            
            # Clean up the extracted text
            extracted_text = extracted_text.replace(prompt, "").strip()
            
            return {
                "success": True,
                "extracted_text": extracted_text,
                "image_path": image_path,
                "model_used": self.model_name,
                "device": self.device
            }
            
        except Exception as e:
            logger.error(f"Error extracting text from image {image_path}: {str(e)}")
            return {
                "success": False,
                "error": str(e),
                "image_path": image_path
            }
    
    def extract_text_with_grounding(self, image_path: str, target_text: str = None) -> Dict[str, Any]:
        """
        Extract text with grounding capabilities (locate specific text)
        
        Args:
            image_path (str): Path to the image file
            target_text (str, optional): Specific text to locate in the image
            
        Returns:
            Dict containing extracted text and location information
        """
        if self.model is None or self.tokenizer is None:
            self.load_model()
        
        try:
            image = Image.open(image_path)
            if image.mode != 'RGB':
                image = image.convert('RGB')
            
            if target_text:
                prompt = f"<|grounding|>Locate <|ref|>{target_text}<|/ref|> in the image."
            else:
                prompt = "<|grounding|>Extract all text from this image with location information."
            
            inputs = self.tokenizer(
                prompt, 
                image, 
                return_tensors="pt"
            ).to(self.device)
            
            with torch.no_grad():
                outputs = self.model.generate(
                    **inputs,
                    max_new_tokens=512,
                    do_sample=False,
                    temperature=0.1,
                    pad_token_id=self.tokenizer.eos_token_id
                )
            
            extracted_text = self.tokenizer.decode(
                outputs[0], 
                skip_special_tokens=True
            )
            
            extracted_text = extracted_text.replace(prompt, "").strip()
            
            return {
                "success": True,
                "extracted_text": extracted_text,
                "grounding_info": target_text if target_text else "all_text",
                "image_path": image_path,
                "model_used": self.model_name
            }
            
        except Exception as e:
            logger.error(f"Error in grounding extraction from {image_path}: {str(e)}")
            return {
                "success": False,
                "error": str(e),
                "image_path": image_path
            }
    
    def convert_to_markdown(self, image_path: str) -> Dict[str, Any]:
        """
        Convert document image to markdown format
        
        Args:
            image_path (str): Path to the image file
            
        Returns:
            Dict containing markdown formatted text
        """
        if self.model is None or self.tokenizer is None:
            self.load_model()
        
        try:
            image = Image.open(image_path)
            if image.mode != 'RGB':
                image = image.convert('RGB')
            
            prompt = "<|grounding|>Convert the document to markdown format."
            
            inputs = self.tokenizer(
                prompt, 
                image, 
                return_tensors="pt"
            ).to(self.device)
            
            with torch.no_grad():
                outputs = self.model.generate(
                    **inputs,
                    max_new_tokens=1024,
                    do_sample=False,
                    temperature=0.1,
                    pad_token_id=self.tokenizer.eos_token_id
                )
            
            markdown_text = self.tokenizer.decode(
                outputs[0], 
                skip_special_tokens=True
            )
            
            markdown_text = markdown_text.replace(prompt, "").strip()
            
            return {
                "success": True,
                "markdown_text": markdown_text,
                "image_path": image_path,
                "model_used": self.model_name
            }
            
        except Exception as e:
            logger.error(f"Error converting to markdown from {image_path}: {str(e)}")
            return {
                "success": False,
                "error": str(e),
                "image_path": image_path
            }

# Global OCR service instance
ocr_service = DeepSeekOCRService()

def get_ocr_service() -> DeepSeekOCRService:
    """
    Get the global OCR service instance
    
    Returns:
        DeepSeekOCRService: The OCR service instance
    """
    return ocr_service