Update app.py

app.py

@@ -18,6 +18,10 @@ from openpyxl.utils import get_column_letter
 from io import BytesIO
 import base64
 import hashlib
+import requests
+import tempfile
+from pathlib import Path
+import mimetypes
 
 # Configure logging
 logging.basicConfig(level=logging.INFO)
@@ -32,6 +36,17 @@ CONFIDENCE_THRESHOLD = 0.65
 BATCH_SIZE = 8  # Reduced batch size for CPU
 MAX_WORKERS = 4  # Number of worker threads for processing
 
+# IMPORTANT: Set PyTorch thread configuration at the module level
+# before any parallel work starts
+if not torch.cuda.is_available():
+    # Set thread configuration only once at the beginning
+    torch.set_num_threads(MAX_WORKERS)
+    try:
+        # Only set interop threads if it hasn't been set already
+        torch.set_num_interop_threads(MAX_WORKERS)
+    except RuntimeError as e:
+        logger.warning(f"Could not set interop threads: {str(e)}")
+
 # Get password hash from environment variable (more secure)
 ADMIN_PASSWORD_HASH = os.environ.get('ADMIN_PASSWORD_HASH')
 
@@ -41,10 +56,138 @@ if not ADMIN_PASSWORD_HASH:
 # Excel file path for logs
 EXCEL_LOG_PATH = "/tmp/prediction_logs.xlsx"
 
+# OCR API settings
+OCR_API_KEY = "9e11346f1288957"  # This is a partial key - replace with the full one
+OCR_API_ENDPOINT = "https://api.ocr.space/parse/image"
+OCR_MAX_PDF_PAGES = 3
+OCR_MAX_FILE_SIZE_MB = 1
+
+# Configure logging for OCR module
+ocr_logger = logging.getLogger("ocr_module")
+ocr_logger.setLevel(logging.INFO)
+
+class OCRProcessor:
+    """
+    Handles OCR processing of image and document files using OCR.space API
+    """
+    def __init__(self, api_key: str = OCR_API_KEY):
+        self.api_key = api_key
+        self.endpoint = OCR_API_ENDPOINT
+        
+    def process_file(self, file_path: str) -> Dict:
+        """
+        Process a file using OCR.space API
+        """
+        start_time = time.time()
+        ocr_logger.info(f"Starting OCR processing for file: {os.path.basename(file_path)}")
+        
+        # Validate file size
+        file_size_mb = os.path.getsize(file_path) / (1024 * 1024)
+        if file_size_mb > OCR_MAX_FILE_SIZE_MB:
+            ocr_logger.warning(f"File size ({file_size_mb:.2f} MB) exceeds limit of {OCR_MAX_FILE_SIZE_MB} MB")
+            return {
+                "success": False,
+                "error": f"File size ({file_size_mb:.2f} MB) exceeds limit of {OCR_MAX_FILE_SIZE_MB} MB",
+                "text": ""
+            }
+        
+        # Determine file type and handle accordingly
+        file_type = self._get_file_type(file_path)
+        ocr_logger.info(f"Detected file type: {file_type}")
+        
+        # Prepare the API request
+        with open(file_path, 'rb') as f:
+            file_data = f.read()
+            
+        # Set up API parameters
+        payload = {
+            'isOverlayRequired': 'false',
+            'language': 'eng',
+            'OCREngine': '2',  # Use more accurate engine
+            'scale': 'true',
+            'detectOrientation': 'true',
+        }
+        
+        # For PDF files, check page count limitations
+        if file_type == 'application/pdf':
+            ocr_logger.info("PDF document detected, enforcing page limit")
+            payload['filetype'] = 'PDF'
+        
+        # Prepare file for OCR API
+        files = {
+            'file': (os.path.basename(file_path), file_data, file_type)
+        }
+        
+        headers = {
+            'apikey': self.api_key,
+        }
+        
+        # Make the OCR API request
+        try:
+            ocr_logger.info("Sending request to OCR.space API")
+            response = requests.post(
+                self.endpoint, 
+                files=files,
+                data=payload,
+                headers=headers
+            )
+            response.raise_for_status()
+            result = response.json()
+            
+            # Process the OCR results
+            if result.get('OCRExitCode') in [1, 2]:  # Success or partial success
+                extracted_text = self._extract_text_from_result(result)
+                processing_time = time.time() - start_time
+                ocr_logger.info(f"OCR processing completed in {processing_time:.2f} seconds")
+                
+                return {
+                    "success": True,
+                    "text": extracted_text,
+                    "word_count": len(extracted_text.split()),
+                    "processing_time_ms": int(processing_time * 1000)
+                }
+            else:
+                ocr_logger.error(f"OCR API error: {result.get('ErrorMessage', 'Unknown error')}")
+                return {
+                    "success": False,
+                    "error": result.get('ErrorMessage', 'OCR processing failed'),
+                    "text": ""
+                }
+                
+        except requests.exceptions.RequestException as e:
+            ocr_logger.error(f"OCR API request failed: {str(e)}")
+            return {
+                "success": False,
+                "error": f"OCR API request failed: {str(e)}",
+                "text": ""
+            }
+    
+    def _extract_text_from_result(self, result: Dict) -> str:
+        """
+        Extract all text from the OCR API result
+        """
+        extracted_text = ""
+        
+        if 'ParsedResults' in result and result['ParsedResults']:
+            for parsed_result in result['ParsedResults']:
+                if parsed_result.get('ParsedText'):
+                    extracted_text += parsed_result['ParsedText']
+        
+        return extracted_text
+    
+    def _get_file_type(self, file_path: str) -> str:
+        """
+        Determine MIME type of a file
+        """
+        mime_type, _ = mimetypes.guess_type(file_path)
+        if mime_type is None:
+            # Default to binary if MIME type can't be determined
+            return 'application/octet-stream'
+        return mime_type
+
 def is_admin_password(input_text: str) -> bool:
     """
     Check if the input text matches the admin password using secure hash comparison.
-    This prevents the password from being visible in the source code.
     """
     # Hash the input text
     input_hash = hashlib.sha256(input_text.strip().encode()).hexdigest()
@@ -105,11 +248,6 @@ class TextWindowProcessor:
 
 class TextClassifier:
     def __init__(self):
-        # Set thread configuration before any model loading or parallel work
-        if not torch.cuda.is_available():
-            torch.set_num_threads(MAX_WORKERS)
-            torch.set_num_interop_threads(MAX_WORKERS)
-            
         self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         self.model_name = MODEL_NAME
         self.tokenizer = None
@@ -253,7 +391,7 @@ class TextClassifier:
                 for window_idx, indices in enumerate(batch_indices):
                     center_idx = len(indices) // 2
                     center_weight = 0.7  # Higher weight for center sentence
-                    edge_weight = 0.3 / (len(indices) - 1)  # Distribute remaining weight
+                    edge_weight = 0.3 / (len(indices) - 1) if len(indices) > 1 else 0  # Distribute remaining weight
 
                     for pos, sent_idx in enumerate(indices):
                         # Apply higher weight to center sentence
@@ -276,10 +414,10 @@ class TextClassifier:
 
                 # Apply minimal smoothing at prediction boundaries
                 if i > 0 and i < len(sentences) - 1:
-                    prev_human = sentence_scores[i-1]['human_prob'] / sentence_appearances[i-1]
-                    prev_ai = sentence_scores[i-1]['ai_prob'] / sentence_appearances[i-1]
-                    next_human = sentence_scores[i+1]['human_prob'] / sentence_appearances[i+1]
-                    next_ai = sentence_scores[i+1]['ai_prob'] / sentence_appearances[i+1]
+                    prev_human = sentence_scores[i-1]['human_prob'] / max(sentence_appearances[i-1], 1e-10)
+                    prev_ai = sentence_scores[i-1]['ai_prob'] / max(sentence_appearances[i-1], 1e-10)
+                    next_human = sentence_scores[i+1]['human_prob'] / max(sentence_appearances[i+1], 1e-10)
+                    next_ai = sentence_scores[i+1]['ai_prob'] / max(sentence_appearances[i+1], 1e-10)
 
                     # Check if we're at a prediction boundary
                     current_pred = 'human' if human_prob > ai_prob else 'ai'
@@ -354,6 +492,72 @@ class TextClassifier:
             'num_sentences': num_sentences
         }
 
+# Function to handle file upload, OCR processing, and text analysis
+def handle_file_upload_and_analyze(file_obj, mode: str, classifier) -> tuple:
+    """
+    Handle file upload, OCR processing, and text analysis
+    """
+    if file_obj is None:
+        return (
+            "No file uploaded",
+            "Please upload a file to analyze",
+            "No file uploaded for analysis"
+        )
+    
+    # Create a temporary file with an appropriate extension based on content
+    content_start = file_obj[:20]  # Look at the first few bytes
+    
+    # Default to .bin extension
+    file_ext = ".bin"
+    
+    # Try to detect PDF files
+    if content_start.startswith(b'%PDF'):
+        file_ext = ".pdf"
+    # For images, detect by common magic numbers
+    elif content_start.startswith(b'\xff\xd8'):  # JPEG
+        file_ext = ".jpg"
+    elif content_start.startswith(b'\x89PNG'):  # PNG
+        file_ext = ".png"
+    elif content_start.startswith(b'GIF'):      # GIF
+        file_ext = ".gif"
+    
+    # Create a temporary file with the detected extension
+    with tempfile.NamedTemporaryFile(delete=False, suffix=file_ext) as temp_file:
+        temp_file_path = temp_file.name
+        # Write uploaded file data to the temporary file
+        temp_file.write(file_obj)
+    
+    try:
+        # Process the file with OCR
+        ocr_processor = OCRProcessor()
+        ocr_result = ocr_processor.process_file(temp_file_path)
+        
+        if not ocr_result["success"]:
+            return (
+                "OCR Processing Error",
+                ocr_result["error"],
+                "Failed to extract text from the uploaded file"
+            )
+        
+        # Get the extracted text
+        extracted_text = ocr_result["text"]
+        
+        # If no text was extracted
+        if not extracted_text.strip():
+            return (
+                "No text extracted",
+                "The OCR process did not extract any text from the uploaded file.",
+                "No text was found in the uploaded file"
+            )
+        
+        # Call the original text analysis function with the extracted text
+        return analyze_text(extracted_text, mode, classifier)
+    
+    finally:
+        # Clean up the temporary file
+        if os.path.exists(temp_file_path):
+            os.remove(temp_file_path)
+
 def initialize_excel_log():
     """Initialize the Excel log file if it doesn't exist."""
     if not os.path.exists(EXCEL_LOG_PATH):
@@ -381,6 +585,7 @@ def initialize_excel_log():
         wb.save(EXCEL_LOG_PATH)
         logger.info(f"Initialized Excel log file at {EXCEL_LOG_PATH}")
 
+
 def log_prediction_data(input_text, word_count, prediction, confidence, execution_time, mode):
     """Log prediction data to an Excel file in the /tmp directory."""
     # Initialize the Excel file if it doesn't exist
@@ -423,6 +628,7 @@ def log_prediction_data(input_text, word_count, prediction, confidence, executio
         logger.error(f"Error logging prediction data to Excel: {str(e)}")
         return False
 
+
 def get_logs_as_base64():
     """Read the Excel logs file and return as base64 for downloading."""
     if not os.path.exists(EXCEL_LOG_PATH):
@@ -441,6 +647,7 @@ def get_logs_as_base64():
         logger.error(f"Error reading Excel logs: {str(e)}")
         return None
 
+
 def analyze_text(text: str, mode: str, classifier: TextClassifier) -> tuple:
     """Analyze text using specified mode and return formatted results."""
     # Check if the input text matches the admin password using secure comparison
@@ -563,47 +770,127 @@ def analyze_text(text: str, mode: str, classifier: TextClassifier) -> tuple:
 # Initialize the classifier globally
 classifier = TextClassifier()
 
-# Create Gradio interface
-demo = gr.Interface(
-    fn=lambda text, mode: analyze_text(text, mode, classifier),
-    inputs=[
-        gr.Textbox(
-            lines=8,
-            placeholder="Enter text to analyze...",
-            label="Input Text"
-        ),
-        gr.Radio(
-            choices=["quick", "detailed"],
-            value="quick",
-            label="Analysis Mode",
-            info="Quick mode for faster analysis, Detailed mode for sentence-level analysis"
+# Create Gradio interface with a small file upload button next to the radio buttons
+def create_interface():
+    # Custom CSS for the interface
+    css = """
+    #analyze-btn {
+        background-color: #FF8C00 !important;
+        border-color: #FF8C00 !important;
+        color: white !important;
+    }
+    
+    .file-upload-btn {
+        margin-left: 10px;
+        background-color: transparent;
+        border: none;
+        cursor: pointer;
+        padding: 0;
+        display: inline-flex;
+        align-items: center;
+        justify-content: center;
+        width: 30px;
+        height: 30px;
+        font-size: 18px;
+    }
+    
+    .file-upload-btn:hover {
+        opacity: 0.8;
+    }
+    """
+    
+    with gr.Blocks(css=css, title="AI Text Detector") as demo:
+        gr.Markdown("# AI Text Detector")
+        gr.Markdown("Analyze text to detect if it was written by a human or AI. Choose between quick scan and detailed sentence-level analysis. 200+ words suggested for accurate predictions.")
+        
+        with gr.Row():
+            # Left column - Input
+            with gr.Column(scale=1):
+                # Text input area
+                text_input = gr.Textbox(
+                    lines=8,
+                    placeholder="Enter text to analyze...",
+                    label="Input Text"
+                )
+                
+                # Analysis mode selection row with file upload button
+                with gr.Row():
+                    mode_selection = gr.Radio(
+                        choices=["quick", "detailed"],
+                        value="quick",
+                        label="Analysis Mode",
+                        info="Quick mode for faster analysis, Detailed mode for sentence-level analysis"
+                    )
+                    
+                    # Hidden file upload component
+                    file_upload = gr.File(
+                        file_types=["image", "pdf", "doc", "docx"],
+                        visible=False,
+                        type="binary"
+                    )
+                    
+                    # Visible file upload button (paperclip icon)
+                    file_button = gr.Button(
+                        "📎", 
+                        elem_classes=["file-upload-btn"],
+                    )
+                
+                # Analyze button with custom styling
+                analyze_btn = gr.Button("Analyze Text", elem_id="analyze-btn")
+            
+            # Right column - Results
+            with gr.Column(scale=1):
+                output_html = gr.HTML(label="Highlighted Analysis")
+                output_sentences = gr.Textbox(label="Sentence-by-Sentence Analysis", lines=10)
+                output_result = gr.Textbox(label="Overall Result", lines=4)
+        
+        # Connect components
+        # 1. Analyze button click
+        analyze_btn.click(
+            fn=lambda text, mode: analyze_text(text, mode, classifier),
+            inputs=[text_input, mode_selection],
+            outputs=[output_html, output_sentences, output_result]
+        )
+        
+        # 2. File button click to trigger file upload
+        file_button.click(
+            fn=lambda: gr.update(visible=True),
+            inputs=None,
+            outputs=file_upload
         )
-    ],
-    outputs=[
-        gr.HTML(label="Highlighted Analysis"),
-        gr.Textbox(label="Sentence-by-Sentence Analysis", lines=10),
-        gr.Textbox(label="Overall Result", lines=4)
-    ],
-    title="AI Text Detector",
-    description="Analyze text to detect if it was written by a human or AI. Choose between quick scan and detailed sentence-level analysis. 200+ words suggested for accurate predictions.",
-    api_name="predict",
-    flagging_mode="never"
-)
-
-# Get the FastAPI app from Gradio
-app = demo.app
-
-# Add CORS middleware
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],  # For development
-    allow_credentials=True,
-    allow_methods=["GET", "POST", "OPTIONS"],
-    allow_headers=["*"],
-)
-
-# Ensure CORS is applied before launching
+        
+        # 3. File upload change event
+        file_upload.change(
+            fn=handle_file_upload_and_analyze,
+            inputs=[file_upload, mode_selection],
+            outputs=[output_html, output_sentences, output_result]
+        )
+    
+    return demo
+
+# Setup the app with CORS middleware
+def setup_app():
+    demo = create_interface()
+    
+    # Get the FastAPI app from Gradio
+    app = demo.app
+    
+    # Add CORS middleware
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],  # For development
+        allow_credentials=True,
+        allow_methods=["GET", "POST", "OPTIONS"],
+        allow_headers=["*"],
+    )
+    
+    return demo
+
+# Initialize the application
 if __name__ == "__main__":
+    demo = setup_app()
+    
+    # Start the server
     demo.queue()
     demo.launch(
         server_name="0.0.0.0",