Update app.py

app.py

@@ -2,11 +2,152 @@
 
 
 
+# import gradio as gr
+# import json
+# import os
+# import tempfile
+# import img2pdf
+# from img2pdf import Rotation
+# from pathlib import Path
+
+# # ==============================
+# # PIPELINE IMPORT
+# # ==============================
+# try:
+#     from working_yolo_pipeline import run_document_pipeline, DEFAULT_LAYOUTLMV3_MODEL_PATH, WEIGHTS_PATH
+# except ImportError:
+#     print("Warning: 'working_yolo_pipeline.py' not found. Using dummy paths.")
+#     def run_document_pipeline(*args):
+#         return {"error": "Placeholder pipeline function called."}
+#     DEFAULT_LAYOUTLMV3_MODEL_PATH = "./models/layoutlmv3_model"
+#     WEIGHTS_PATH = "./weights/yolo_weights.pt"
+
+# def process_file(uploaded_files, layoutlmv3_model_path=None):
+#     """
+#     Robust handler for multiple or single file uploads.
+#     """
+#     if uploaded_files is None:
+#         return "❌ Error: No files uploaded.", None
+
+#     # --- THE ROBUST FIX ---
+#     # Gradio sometimes sends a single dict even when set to multiple.
+#     # We force everything into a list so the rest of the logic doesn't break.
+#     if not isinstance(uploaded_files, list):
+#         file_list = [uploaded_files]
+#     else:
+#         file_list = uploaded_files
+
+#     if len(file_list) == 0:
+#         return "❌ Error: Empty file list.", None
+#     # ----------------------
+
+#     # 1. Resolve all file paths safely
+#     resolved_paths = []
+#     for f in file_list:
+#         try:
+#             if isinstance(f, dict) and "path" in f:
+#                 resolved_paths.append(f["path"])
+#             elif hasattr(f, 'path'):
+#                 resolved_paths.append(f.path)
+#             else:
+#                 resolved_paths.append(str(f))
+#         except Exception as e:
+#             print(f"Error resolving path for {f}: {e}")
+
+#     if not resolved_paths:
+#         return "❌ Error: Could not resolve file paths.", None
+
+#     # 2. Determine if we should merge into a single PDF
+#     first_file = Path(resolved_paths[0])
+#     is_image = first_file.suffix.lower() in ['.jpg', '.jpeg', '.png', '.bmp', '.webp', '.tiff']
+    
+#     try:
+#         # If it's multiple files or just one image, wrap it in a PDF
+#         if len(resolved_paths) > 1 or is_image:
+#             print(f"📦 Converting {len(resolved_paths)} image(s) to a single PDF...")
+#             temp_pdf = tempfile.NamedTemporaryFile(delete=False, suffix=".pdf")
+#             with open(temp_pdf.name, "wb") as f_out:
+#                 # f_out.write(img2pdf.convert(resolved_paths))
+#                 f_out.write(img2pdf.convert(resolved_paths, rotation=Rotation.ifvalid))
+
+#             processing_path = temp_pdf.name
+#         else:
+#             # It's a single PDF
+#             processing_path = resolved_paths[0]
+
+#         # 3. Standard Pipeline Checks
+#         final_model_path = layoutlmv3_model_path or DEFAULT_LAYOUTLMV3_MODEL_PATH
+#         if not os.path.exists(final_model_path):
+#             return f"❌ Error: Model not found at {final_model_path}", None
+
+#         # 4. Call the pipeline
+#         print(f"🚀 Starting pipeline for: {processing_path}")
+#         result = run_document_pipeline(processing_path, final_model_path)
+
+#         if result is None:
+#             return "❌ Error: Pipeline returned None.", None
+
+#         # 5. Prepare output
+#         temp_output = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.json', prefix='analysis_')
+#         with open(temp_output.name, 'w', encoding='utf-8') as f:
+#             json.dump(result, f, indent=2, ensure_ascii=False)
+
+#         return json.dumps(result, indent=2, ensure_ascii=False), temp_output.name
+
+#     except Exception as e:
+#         import traceback
+#         traceback.print_exc()
+#         return f"❌ Error: {str(e)}", None
+
+# # ==============================
+# # GRADIO INTERFACE
+# # ==============================
+# with gr.Blocks(title="Document Analysis Pipeline") as demo:
+
+#     gr.Markdown("# 📄 Document & Image Analysis Pipeline")
+
+#     with gr.Row():
+#         with gr.Column(scale=1):
+#             file_input = gr.File(
+#                 label="Upload PDFs or Images",
+#                 file_types=[".pdf", ".jpg", ".jpeg", ".png", ".bmp", ".webp", ".tiff"],
+#                 file_count="multiple", # Keep this
+#                 type="filepath"       # Keep this
+#             )
+
+#             model_path_input = gr.Textbox(
+#                 label="Model Path",
+#                 value=DEFAULT_LAYOUTLMV3_MODEL_PATH
+#             )
+
+#             process_btn = gr.Button("🚀 Process Files", variant="primary")
+
+#         with gr.Column(scale=2):
+#             json_output = gr.Code(label="JSON Output", language="json", lines=20)
+#             download_output = gr.File(label="Download JSON")
+
+#     process_btn.click(
+#         fn=process_file,
+#         inputs=[file_input, model_path_input],
+#         outputs=[json_output, download_output]
+#     )
+
+# if __name__ == "__main__":
+#     demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)
+
+
+
+
+
+
+
+
 import gradio as gr
 import json
 import os
 import tempfile
 import img2pdf
+import glob
 from img2pdf import Rotation
 from pathlib import Path
 
@@ -25,6 +166,8 @@ except ImportError:
 def process_file(uploaded_files, layoutlmv3_model_path=None):
     """
     Robust handler for multiple or single file uploads.
+    Returns the final JSON and the file path for download.
+    If the pipeline fails at BIO conversion, it attempts to return the raw predictions for debugging.
     """
     if uploaded_files is None:
         return "❌ Error: No files uploaded.", None
@@ -67,9 +210,7 @@ def process_file(uploaded_files, layoutlmv3_model_path=None):
             print(f"📦 Converting {len(resolved_paths)} image(s) to a single PDF...")
             temp_pdf = tempfile.NamedTemporaryFile(delete=False, suffix=".pdf")
             with open(temp_pdf.name, "wb") as f_out:
-                # f_out.write(img2pdf.convert(resolved_paths))
                 f_out.write(img2pdf.convert(resolved_paths, rotation=Rotation.ifvalid))
-
             processing_path = temp_pdf.name
         else:
             # It's a single PDF
@@ -84,10 +225,33 @@ def process_file(uploaded_files, layoutlmv3_model_path=None):
         print(f"🚀 Starting pipeline for: {processing_path}")
         result = run_document_pipeline(processing_path, final_model_path)
 
-        if result is None:
-            return "❌ Error: Pipeline returned None.", None
+        # --- DEBUGGING LOGIC FOR STEP 3 FAILURE ---
+        if result is None or (isinstance(result, list) and len(result) == 0):
+            print("⚠️ Pipeline returned no structured data. Looking for raw predictions for debugging...")
+            
+            # Based on your logs, the pipeline creates a folder like /tmp/pipeline_run_[filename]
+            base_name = Path(processing_path).stem
+            search_pattern = f"/tmp/pipeline_run_{base_name}*/*_raw_predictions.json"
+            possible_files = glob.glob(search_pattern)
+
+            if possible_files:
+                debug_file = possible_files[0]
+                print(f"🔍 DEBUG: Found raw predictions at {debug_file}")
+                with open(debug_file, 'r', encoding='utf-8') as f:
+                    raw_data = json.load(f)
+                
+                # Return the raw labels to the UI so you can see why it failed
+                return (
+                    "⚠️ WARNING: BIO Decoding Failed (Step 3).\n"
+                    "Showing RAW LayoutLMv3 predictions instead for analysis:\n\n" + 
+                    json.dumps(raw_data, indent=2, ensure_ascii=False), 
+                    debug_file
+                )
+            
+            return "❌ Error: Pipeline failed and no intermediate raw prediction file was found.", None
+        # ------------------------------------------
 
-        # 5. Prepare output
+        # 5. Prepare output (Successful Path)
         temp_output = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.json', prefix='analysis_')
         with open(temp_output.name, 'w', encoding='utf-8') as f:
             json.dump(result, f, indent=2, ensure_ascii=False)
@@ -105,14 +269,15 @@ def process_file(uploaded_files, layoutlmv3_model_path=None):
 with gr.Blocks(title="Document Analysis Pipeline") as demo:
 
     gr.Markdown("# 📄 Document & Image Analysis Pipeline")
+    gr.Markdown("### 🛠 Debug Mode Active: If Step 3 fails, the Raw Prediction file will be returned.")
 
     with gr.Row():
         with gr.Column(scale=1):
             file_input = gr.File(
                 label="Upload PDFs or Images",
                 file_types=[".pdf", ".jpg", ".jpeg", ".png", ".bmp", ".webp", ".tiff"],
-                file_count="multiple", # Keep this
-                type="filepath"       # Keep this
+                file_count="multiple",
+                type="filepath"
             )
 
             model_path_input = gr.Textbox(
@@ -123,8 +288,8 @@ with gr.Blocks(title="Document Analysis Pipeline") as demo:
             process_btn = gr.Button("🚀 Process Files", variant="primary")
 
         with gr.Column(scale=2):
-            json_output = gr.Code(label="JSON Output", language="json", lines=20)
-            download_output = gr.File(label="Download JSON")
+            json_output = gr.Code(label="JSON Output (Structured or Raw Predictions)", language="json", lines=20)
+            download_output = gr.File(label="Download JSON File")
 
     process_btn.click(
         fn=process_file,
@@ -133,4 +298,5 @@ with gr.Blocks(title="Document Analysis Pipeline") as demo:
     )
 
 if __name__ == "__main__":
+    # Note: 0.0.0.0 allows access from outside the container/host
     demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)