Update app.py

app.py

@@ -1,15 +1,37 @@
 import base64
 import os
-
-app_data_path = '/app/app_data'
+import tempfile
+
+# Use temporary directory or current working directory for app data
+# This avoids permission issues in containerized environments
+try:
+    # Try to use /tmp first (usually writable in containers)
+    app_data_path = '/tmp/app_data'
+    os.makedirs(app_data_path, exist_ok=True)
+except PermissionError:
+    # Fallback to current directory or temp directory
+    app_data_path = os.path.join(os.getcwd(), 'app_data')
+    try:
+        os.makedirs(app_data_path, exist_ok=True)
+    except PermissionError:
+        # Last resort: use system temp directory
+        app_data_path = tempfile.mkdtemp(prefix='app_data_')
 
 # Make sure the necessary directories exist and have proper permissions
-os.makedirs(f'{app_data_path}/ultralytics', exist_ok=True)
-os.makedirs(f'{app_data_path}/.keras', exist_ok=True)
-
-# Set environment variables for the respective paths
-os.environ['ULTRALYTICS_CONFIG_DIR'] = f'{app_data_path}/ultralytics'
-os.environ['KERAS_HOME'] = f'{app_data_path}/.keras'
+try:
+    os.makedirs(f'{app_data_path}/ultralytics', exist_ok=True)
+    os.makedirs(f'{app_data_path}/.keras', exist_ok=True)
+    
+    # Set environment variables for the respective paths
+    os.environ['ULTRALYTICS_CONFIG_DIR'] = f'{app_data_path}/ultralytics'
+    os.environ['KERAS_HOME'] = f'{app_data_path}/.keras'
+    
+    print(f"Successfully created app data directory at: {app_data_path}")
+except PermissionError as e:
+    print(f"Warning: Could not create app data directories: {e}")
+    # Set environment variables to current directory as fallback
+    os.environ['ULTRALYTICS_CONFIG_DIR'] = os.getcwd()
+    os.environ['KERAS_HOME'] = os.getcwd()
 
 import cv2
 import numpy as np
@@ -51,15 +73,18 @@ def create_cnn_embedding_model():
         Dense(512, activation='linear')  # Final dense layer to produce a 512-dimensional embedding
     ])
     return model
+
 cnn_model = create_cnn_embedding_model()
 cnn_model.compile(optimizer='adam', loss='mse') 
 # Using MSE loss as we are not training for classification
+
 def cosine(embedding1, embedding2):
     dot_product = np.dot(embedding1, embedding2)
     norm1 = np.linalg.norm(embedding1)
     norm2 = np.linalg.norm(embedding2)
     similarity = dot_product / (norm1 * norm2)
     return similarity
+
 @app.route('/CNN-login', methods=['POST'])
 def cnnlogin():
     # Check for uploaded image
@@ -207,7 +232,19 @@ def register():
     # Calculate mean embeddings
     mean_facenet_embedding = np.mean(facenet_embeddings, axis=0).astype(float).tolist()
     mean_cnn_embedding = np.mean(cnn_embeddings, axis=0).astype(float).tolist()
-    cnn_model.save_weights('cnn_model.weights.h5') 
+    
+    # Save model weights with error handling
+    try:
+        weights_path = os.path.join(app_data_path, 'cnn_model.weights.h5')
+        cnn_model.save_weights(weights_path)
+    except Exception as e:
+        print(f"Warning: Could not save model weights: {e}")
+        # Try saving in current directory
+        try:
+            cnn_model.save_weights('cnn_model.weights.h5')
+        except Exception as e2:
+            print(f"Warning: Could not save model weights in current directory: {e2}")
+    
     # Create user data
     id = mongo.db.data.count_documents({}) + 1
     user_data = {
@@ -251,20 +288,38 @@ def load_embeddings_from_db():
 
 # Load face embeddings from MongoDB initially
 face_data, labels, names = load_embeddings_from_db()
-cnn_model.load_weights('cnn_model.weights.h5')
-
 
+# Load CNN model weights with error handling
+try:
+    weights_path = os.path.join(app_data_path, 'cnn_model.weights.h5')
+    if os.path.exists(weights_path):
+        cnn_model.load_weights(weights_path)
+        print("Loaded CNN model weights from app_data directory")
+    elif os.path.exists('cnn_model.weights.h5'):
+        cnn_model.load_weights('cnn_model.weights.h5')
+        print("Loaded CNN model weights from current directory")
+    else:
+        print("No existing CNN model weights found - will use initialized weights")
+except Exception as e:
+    print(f"Warning: Could not load CNN model weights: {e}")
 
 # Reload embeddings to update after a new registration
 def reload_embeddings():
     global face_data, labels, names
-    cnn_model.load_weights('cnn_model.weights.h5')
+    try:
+        weights_path = os.path.join(app_data_path, 'cnn_model.weights.h5')
+        if os.path.exists(weights_path):
+            cnn_model.load_weights(weights_path)
+        elif os.path.exists('cnn_model.weights.h5'):
+            cnn_model.load_weights('cnn_model.weights.h5')
+    except Exception as e:
+        print(f"Warning: Could not reload CNN model weights: {e}")
+    
     face_data, labels, names = load_embeddings_from_db()
 
 # Recognize faces using MongoDB-stored embeddings
 model = YOLO('yolov5s.pt')  # Replace with your YOLO model path
 
-
 @app.route('/crowd', methods=['POST'])
 def upload_image():
     if 'image' not in request.files:
@@ -397,3 +452,5 @@ def get_attendance():
     records = list(mongo.db.attendance1.find({}, {"_id": 0}))
     return jsonify({"attendance": records})
 
+if __name__ == '__main__':
+    app.run(debug=True)