Update app.py

app.py

@@ -57,6 +57,11 @@ def process_image(input_image):
     # Step 4: Crop the image using the bounding box
     cropped_image = enhanced_image.crop((xmin, ymin, xmax, ymax))
     
+    # OPTIMIZATION: Resize the image to reduce processing time
+    # Calculate aspect ratio to maintain proportions
+    max_size = (800, 800)  # Reduced from original size
+    cropped_image.thumbnail(max_size, Image.LANCZOS)
+    
     # Select the corresponding OCR prompt based on the YOLO label
     if label.lower() == "front":
         doc_prompt = front
@@ -93,12 +98,14 @@ def process_image(input_image):
     )
     inputs = {k: v.to(device) for k, v in inputs.items()}
     
+    # OPTIMIZATION: Modified generation parameters for faster processing
     output = ocr_model.generate(
         **inputs,
-        temperature=0.8,
-        max_new_tokens=50,
+        temperature=0.2,  # Reduced from 0.8 to 0.2 for faster, more deterministic output
+        max_new_tokens=40,  # Slightly reduced from 50
         num_return_sequences=1,
-        do_sample=True,
+        do_sample=False,  # Changed to deterministic for speed
+        early_stopping=True  # Add early stopping to prevent unnecessary generation
     )
     prompt_length = inputs["input_ids"].shape[1]
     new_tokens = output[:, prompt_length:]
@@ -124,4 +131,5 @@ iface = gr.Interface(
     ),
 )
 
-iface.launch()
+# Enable queue for better handling of processing time
+iface.launch(share=True)