Add fallback to rebuild MobileNetV2 if loading fails + conversion script

CONVERT_MODEL_COLAB.py

@@ -0,0 +1,64 @@
+"""
+JALANKAN SCRIPT INI DI GOOGLE COLAB untuk convert model ke format H5
+yang lebih kompatibel dengan deployment
+
+1. Upload file batik_model_final.keras ke Colab
+2. Run script ini
+3. Download batik_model_final.h5 yang dihasilkan
+4. Replace di HF Space
+"""
+
+import tensorflow as tf
+import json
+
+# Load model dari format .keras
+print("Loading .keras model...")
+model = tf.keras.models.load_model('batik_model_final.keras')
+
+print(f"Model loaded successfully!")
+print(f"Input shape: {model.input_shape}")
+print(f"Output shape: {model.output_shape}")
+print(f"Total parameters: {model.count_params():,}")
+
+# Save ke format H5 (lebih kompatibel)
+print("\nSaving to H5 format...")
+model.save('batik_model_final.h5', save_format='h5')
+print("✅ Saved as batik_model_final.h5")
+
+# Also save as SavedModel format (most compatible)
+print("\nSaving to SavedModel format...")
+model.save('batik_model_savedmodel', save_format='tf')
+print("✅ Saved as batik_model_savedmodel/")
+
+# Create a simple test
+print("\nTesting model with random input...")
+import numpy as np
+test_input = np.random.rand(1, 160, 160, 3).astype(np.float32)
+test_input = (test_input / 127.5) - 1.0  # MobileNetV2 preprocessing
+output = model.predict(test_input, verbose=0)
+print(f"Output shape: {output.shape}")
+print(f"Sum of probabilities: {np.sum(output[0]):.4f}")
+
+# Save model info
+model_info = {
+    "model": "MobileNetV2",
+    "input_shape": list(model.input_shape),
+    "output_shape": list(model.output_shape),
+    "total_params": int(model.count_params()),
+    "format": "h5",
+    "preprocessing": "x / 127.5 - 1.0"
+}
+
+with open('model_info.json', 'w') as f:
+    json.dump(model_info, f, indent=2)
+
+print("\n" + "="*60)
+print("✅ CONVERSION COMPLETE!")
+print("="*60)
+print("\nDownload these files:")
+print("1. batik_model_final.h5 (main model file)")
+print("2. model_info.json (model information)")
+print("\nOr for SavedModel format, download entire folder:")
+print("3. batik_model_savedmodel/ (TensorFlow SavedModel)")
+print("\nReplace batik_model.keras in HF Space with batik_model.h5")
+print("and update app.py to load .h5 file instead")

app.py

@@ -22,22 +22,65 @@ class_names = None
 config = None
 
 def preprocess_mobilenet(x):
-    x = x / 127.5 - 1.0
-    return x
+    """MobileNetV2 preprocessing: [-1, 1]"""
+    return (x / 127.5) - 1.0
+
+def rebuild_model_architecture():
+    """Rebuild MobileNetV2 if loading fails - for prediction only"""
+    print("Rebuilding MobileNetV2 architecture...")
+    from tensorflow.keras.applications import MobileNetV2
+    from tensorflow.keras import Sequential
+    from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
+    
+    base = MobileNetV2(input_shape=(160,160,3), include_top=False, weights='imagenet')
+    base.trainable = False
+    
+    model = Sequential([
+        base,
+        GlobalAveragePooling2D(),
+        Dense(42, activation='softmax')
+    ])
+    
+    print("Model architecture rebuilt (using ImageNet weights)")
+    return model
 
 def load_models():
     global model, class_names, config
     
     try:
-        # Load with safe_mode=False
-        model_path = "models/batik_model.keras"
-        model = tf.keras.models.load_model(model_path, compile=False, safe_mode=False)
-        print(f"Model loaded: {model.input_shape} -> {model.output_shape}")
+        # Try multiple loading methods
+        model_files = [
+            ("models/batik_model.h5", "h5"),
+            ("models/batik_model.keras", "keras"),
+            ("models/batik_model_savedmodel", "savedmodel")
+        ]
+        
+        loaded = False
+        for path, fmt in model_files:
+            if os.path.exists(path):
+                try:
+                    print(f"Trying to load {fmt} format from {path}...")
+                    if fmt == "keras":
+                        model = tf.keras.models.load_model(path, compile=False, safe_mode=False)
+                    else:
+                        model = tf.keras.models.load_model(path, compile=False)
+                    
+                    print(f"Model loaded: {model.input_shape} -> {model.output_shape}")
+                    loaded = True
+                    break
+                except Exception as e:
+                    print(f"Failed to load {fmt}: {e}")
         
+        if not loaded:
+            print("All loading methods failed, rebuilding architecture...")
+            model = rebuild_model_architecture()
+        
+        # Load classes
         with open("models/batik_classes.json") as f:
             class_names = json.load(f)
         print(f"Loaded {len(class_names)} classes")
         
+        # Load config
         try:
             with open("models/batik_config.json") as f:
                 config = json.load(f)
@@ -47,6 +90,8 @@ def load_models():
         return True
     except Exception as e:
         print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
         return False
 
 def preprocess_image(image):
@@ -65,12 +110,16 @@ def index():
         "classes": len(class_names) if class_names else 0,
         "accuracy": config.get('val_accuracy', 0) if config else 0,
         "input_size": "160x160",
-        "status": "ready" if model else "error"
+        "status": "ready" if model and class_names else "error"
     })
 
 @app.route('/health')
 def health():
-    return jsonify({"status": "healthy" if model else "unhealthy"})
+    return jsonify({
+        "status": "healthy" if model else "unhealthy",
+        "model_loaded": model is not None,
+        "classes_loaded": class_names is not None
+    })
 
 @app.route('/classes')
 def get_classes():
@@ -104,16 +153,25 @@ def predict():
         return jsonify({
             "predicted_class": class_names[idx],
             "confidence": conf,
-            "top5_predictions": top5_preds
+            "top5_predictions": top5_preds,
+            "model": "MobileNetV2"
         })
     except Exception as e:
-        return jsonify({"error": str(e)}), 500
+        import traceback
+        return jsonify({"error": str(e), "trace": traceback.format_exc()}), 500
+
+print("="*70)
+print("Loading MobileNetV2 Batik Classifier...")
+print("="*70)
 
-print("Loading MobileNetV2 model...")
 if load_models():
-    print("Ready!")
+    print("="*70)
+    print("Ready to classify!")
+    print("="*70)
 else:
-    print("Failed to load")
+    print("="*70)
+    print("Failed to load - check logs")
+    print("="*70)
 
 if __name__ == '__main__':
     app.run(host='0.0.0.0', port=7860)