clean model final changes

app.py

@@ -45,6 +45,7 @@ from transformers import (
 )
 from matplotlib import cm
 import gradio as gr
+import tempfile
 
 # ----------------------------------------------------------------------
 # Configuration
@@ -789,7 +790,15 @@ def run_ocr(
     plain_text = re.sub(r"<[^>]*>", "", predicted_html.replace("<br>", "\n"))
     # Write temporary files
 
-    return overlay_img, predicted_html, df_all
+    # return overlay_img, predicted_html
+    # Save plain text to a temporary .txt file
+    txt_dir = tempfile.gettempdir()
+    txt_path = os.path.join(txt_dir, "predictions.txt")
+    with open(txt_path, "w", encoding="utf-8") as f:
+        f.write(plain_text)
+
+
+    return overlay_img, predicted_html, txt_path
 
 
 # ----------------------------------------------------------------------
@@ -797,8 +806,23 @@ def run_ocr(
 #
 def create_gradio_interface():
     """Create and return the Gradio Blocks interface."""
-    with gr.Blocks(title="Old Nepali HTR") as demo:
-        gr.Markdown("""# Old Nepali HTR (Gradio)\n\nUpload a scanned image and (optionally) a segmentation XML file.  Choose preprocessing\nsteps and a highlight metric, then click **Run OCR** to extract the text.\nUncertain tokens are highlighted with tooltips showing alternative predictions.\nYou can edit the plain text below and download it or the full token scores.""")
+    with gr.Blocks(title="Handwritten Text Recognition (Old Nepali)") as demo:
+        gr.Markdown("""# Handwritten Text Recognition (Old Nepali) \n\nUpload an image and (optionally) a segmentation XML file.  Then click **Run OCR** to extract the text.""")
+        gr.HTML("""
+            <style>
+            #prediction-box {
+                border: 1px solid #ccc;
+                padding: 16px;
+                border-radius: 8px;
+                background-color: #f9f9f9;
+                font-size: 18px;
+                line-height: 1.6;
+                min-height: 100px;
+            }
+            
+            }
+            </style>
+            """)
         with gr.Row():
             image_input = gr.Image(type="numpy", label="Upload Image")
             # When used as an input, gr.File returns either a file path or bytes
@@ -811,18 +835,34 @@ def create_gradio_interface():
                 type="binary",
                 file_types=[".xml"],
             )
-        with gr.Row():
-            apply_gray_checkbox = gr.Checkbox(label="Convert to Grayscale", value=False)
-            apply_bin_checkbox = gr.Checkbox(label="Binarize", value=False)
-            metric_radio = gr.Radio([
-                "Relative Probability",
-                "Entropy",
-            ], label="Highlight tokens by", value="Relative Probability")
+        # with gr.Row():
+        #     apply_gray_checkbox = gr.Checkbox(label="Convert to Grayscale", value=False)
+        #     apply_bin_checkbox = gr.Checkbox(label="Binarize", value=False)
+        #     metric_radio = gr.Radio([
+        #         "Relative Probability",
+        #         "Entropy",
+        #     ], label="Highlight tokens by", value="Relative Probability")
         run_btn = gr.Button("Run OCR")
         # Outputs
-        overlay_output = gr.Image(label="Detected Regions")
-        predictions_output = gr.HTML(label="Predictions (HTML)")
-        df_output = gr.DataFrame(label="Token Scores", interactive=False)
+        # overlay_output = gr.Image(label="Detected Regions")
+        # # predictions_output = gr.HTML(label="Predictions (HTML)")
+        # predictions_output = gr.HTML(
+        #     label="Predictions (HTML)",
+        #     elem_id="prediction-box"
+        # )
+        # df_output = gr.DataFrame(label="Token Scores", interactive=False)
+        with gr.Row():
+            with gr.Column(scale=2):
+                overlay_output = gr.Image(label="Detected Regions")
+
+            with gr.Column(scale=2):
+                predictions_output = gr.HTML(
+                    label="Predictions (HTML)",
+                    elem_id="prediction-box"
+                )
+                
+                # df_output = gr.DataFrame(label="Token Scores", interactive=False)
+        
         # txt_file_output = gr.File(label="Download OCR Prediction (.txt)")
         # csv_file_output = gr.File(label="Download Token Scores (.csv)")
         # Editable text
@@ -830,17 +870,18 @@ def create_gradio_interface():
             label="Edit full predicted text", lines=8, interactive=True
         )
         # download_edited_btn = gr.Button("Download edited text")
-
+        txt_file_output = gr.File(label="Download OCR Prediction (.txt)")
         # Callback for OCR
-        def on_run(image, xml, gray, binarize, metric):
-            return run_ocr(image, xml, gray, binarize, metric)
+        def on_run(image, xml):
+            return run_ocr(image, xml, False, False, "Relative Probability")
+
 
         run_btn.click(
             fn=on_run,
-            inputs=[image_input, xml_input, apply_gray_checkbox, apply_bin_checkbox, metric_radio],
-            outputs=[overlay_output, predictions_output, df_output],
+            # inputs=[image_input, xml_input, apply_gray_checkbox, apply_bin_checkbox, metric_radio],
+            inputs=[image_input, xml_input],
+            outputs=[overlay_output, predictions_output, txt_file_output],
         )
-
         # Populate editable text with plain text from predictions
         def update_edited_text(pred_html):
             plain = re.sub(r"<[^>]*>", "", (pred_html or "").replace("<br>", "\n"))