Update app.py

app.py

@@ -6,7 +6,7 @@ def preprocess_image(image):
   bin_image = cv.copyMakeBorder(bin_image, int(0.10 * image.shape[0]), int(0.05 * image.shape[0]), int(0.05 * image.shape[1]), int(0.10 * image.shape[1]), cv.BORDER_CONSTANT, value=(255, 255, 255))
   return bin_image
 
-bin_image = preprocess_image(image)
+#bin_image = preprocess_image(image)
 
 def split_image_into_lines(image):
   lines = []
@@ -42,7 +42,7 @@ def split_image_into_lines(image):
 
   return lines
 
-lines = split_image_into_lines(bin_image)
+#lines = split_image_into_lines(bin_image)
 
 
 def generate_text(line):
@@ -51,24 +51,28 @@ def generate_text(line):
   generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
   return generated_text
 
-with ProcessPoolExecutor() as executor:
-    results = ' '.join(executor.map(generate_text, lines))
-    #improve results with llm
 
-    client = OpenAI()
+def get_improved_result(lines):
+    with ProcessPoolExecutor() as executor:
+        results = ' '.join(executor.map(generate_text, lines))
+        #improve results with llm
+    
+        client = OpenAI()
+    
+        completion = client.chat.completions.create(
+            model="gpt-4o",
+            messages=[
+                {
+                    "role": "user",
+    
+                    "content": f"I have a string that was extracted from an image of handwritten text. The extraction process introduced minor grammatical, spelling, and punctuation errors. Please carefully review the text below and make any necessary corrections to improve readability and accuracy while preserving the original meaning. Do not change the content or style beyond necessary corrections. Return the corrected text only without adding any headings, explanations, or extra formatting. Text: {results}"
+                }
+            ]
+        )
+    
+        improved_text = completion.choices[0].message.content
 
-    completion = client.chat.completions.create(
-        model="gpt-4o",
-        messages=[
-            {
-                "role": "user",
-
-                "content": f"I have a string that was extracted from an image of handwritten text. The extraction process introduced minor grammatical, spelling, and punctuation errors. Please carefully review the text below and make any necessary corrections to improve readability and accuracy while preserving the original meaning. Do not change the content or style beyond necessary corrections. Return the corrected text only without adding any headings, explanations, or extra formatting. Text: {results}"
-            }
-        ]
-    )
-
-    improved_text = completion.choices[0].message.content
+    return improved_text
 
 
 
@@ -105,4 +109,21 @@ max_width = 0.9
 out_image_width = 1500
 top_margin = 100
 
-out_image = put_text(improved_text, font, font_scale, color, thickness, max_width, out_image_width, top_margin)
+#out_image = put_text(improved_text, font, font_scale, color, thickness, max_width, out_image_width, top_margin)
+
+def predict(input_img):
+    bin_image = preprocess_image(input_img)
+    lines = split_image_into_lines(bin_image)
+    improved_text = get_improved_result(lines)
+    out_image = put_text(improved_text, font, font_scale, color, thickness, max_width, out_image_width, top_margin)
+    return out_img
+
+gradio_app = gr.Interface(
+    predict,
+    inputs=gr.Image(label="Image with handwritten text", sources=['upload']),
+    outputs=[gr.Image(label="Output Image")],
+    title="Extract Handwritten Text",
+)
+
+if __name__ == "__main__":
+    gradio_app.launch()