Fixed the inaccurate prediction

app.py

@@ -1,3 +1,10 @@
+# import warnings
+
+# # Ignore SourceChangeWarning
+# warnings.filterwarnings("ignore", category=DeprecationWarning)
+# warnings.filterwarnings("ignore", category=SourceChangeWarning)
+
+# Dependencies
 from flask import Flask, request, render_template, jsonify, send_file, redirect, url_for, flash, send_from_directory, session
 from PIL import Image, ImageDraw
 import torch
@@ -29,11 +36,7 @@ from Layoutlmv3_inference.ocr import prepare_batch_for_inference
 from Layoutlmv3_inference.inference_handler import handle
 import logging
 import os
-import warnings
-
-# Ignore SourceChangeWarning
-warnings.filterwarnings("ignore", category=DeprecationWarning)
-# warnings.filterwarnings("ignore", category=SourceChangeWarning)
+import copy
 
 
 UPLOAD_FOLDER = 'static/temp/uploads'
@@ -104,8 +107,8 @@ def make_predictions(image_paths):
     temp = None
     try:
         # For Windows OS
-        # temp = pathlib.PosixPath  # Save the original state
-        # pathlib.PosixPath = pathlib.WindowsPath  # Change to WindowsPath temporarily
+        temp = pathlib.PosixPath  # Save the original state
+        pathlib.PosixPath = pathlib.WindowsPath  # Change to WindowsPath temporarily
         
         model_path = Path(r'model/export')
         learner = load_learner(model_path)
@@ -129,10 +132,10 @@ def make_predictions(image_paths):
     except Exception as e:
         return {"error in make_predictions": str(e)}
         
-    # finally:
-    #     pathlib.PosixPath = temp 
+    finally:
+        pathlib.PosixPath = temp 
+
 
-import copy
 @app.route('/predict/<filenames>', methods=['GET', 'POST'])
 def predict_files(filenames):
     prediction_results = []
@@ -162,17 +165,19 @@ def predict_files(filenames):
         if os.path.exists(file_path):
             # Call make_predictions automatically
             prediction_result = make_predictions([file_path])  # Pass file_path as a list
-            prediction_results.extend(prediction_result)
-            
-            # Create a copy of prediction_results before deletion
+            prediction_results.append(prediction_result[0])  # Append only the first prediction result
             prediction_results_copy = copy.deepcopy(prediction_results)
-            
-            if prediction_result[0] != 'non-receipt':  # Check if prediction is not 'non-receipt'
-                prediction_results.extend(prediction_result)  # Use extend to add elements of list to another list
-            else:
-                # Delete the image if it's predicted as non-receipt
-                os.remove(file_path)
-            print(image_paths)
+
+            non_receipt_indices = []
+            for i, prediction in enumerate(prediction_results):
+                if prediction == 'non-receipt':
+                    non_receipt_indices.append(i)
+
+            # Delete images in reverse order to avoid index shifting
+            for index in non_receipt_indices[::-1]:
+                file_to_remove = os.path.join('static', 'temp', 'uploads', image_paths[index])
+                if os.path.exists(file_to_remove):
+                    os.remove(file_to_remove)
 
     return render_template('extractor.html', image_paths=image_paths, prediction_results = prediction_results, predictions=dict(zip(image_paths, prediction_results_copy)))