Update app.py

Replaced character-level decoding with direct mapping from model output index to drug names

app.py

@@ -9,6 +9,30 @@ from PIL import Image
 # Load the trained model
 model = tf.keras.models.load_model("prescription_classification_model.keras")
 
+# Define the 78 drug names in the exact order corresponding to your model's output classes
+CLASS_NAMES = [
+    "Beklo", "Maxima", "Leptic", "Esoral", "Omastin", "Esonix", "Canazole", "Fixal",
+    "Progut", "Diflu", "Montair", "Flexilax", "Maxpro", "Vifas", "Conaz", "Fexofast",
+    "Fenadin", "Telfast", "Dinafex", "Ritch", "Renova", "Flugal", "Axodin", "Sergel",
+    "Nexum", "Opton", "Nexcap", "Fexo", "Montex", "Exium", "Lumona", "Napa", "Azithrocin",
+    "Atrizin", "Monas", "Nidazyl", "Metsina", "Baclon", "Rozith", "Bicozin", "Ace",
+    "Amodis", "Alatrol", "Napa Extend", "Rivotril", "Montene", "Filmet", "Aceta",
+    "Tamen", "Bacmax", "Disopan", "Rhinil", "Flamyd", "Metro", "Zithrin", "Candinil",
+    "Lucan-R", "Backtone", "Bacaid", "Etizin", "Az", "Romycin", "Azyth", "Cetisoft",
+    "Dancel", "Tridosil", "Nizoder", "Ketoral", "Ketocon", "Ketotab", "Ketozol",
+    "Denixil", "Provair", "Odmon", "Baclofen", "MKast", "Trilock", "Flexibac"
+]
+
+def decode_prediction(prediction):
+    """
+    Expects prediction to be a numpy array of shape (1, 78).
+    It returns the drug name corresponding to the highest probability.
+    """
+    # Get the index of the highest probability class
+    predicted_index = np.argmax(prediction, axis=-1)[0]
+    # Return the corresponding drug name
+    return CLASS_NAMES[predicted_index]
+
 # Function to preprocess the uploaded image
 def preprocess_image(image):
     image = image.convert("RGB")  # Convert to grayscale
@@ -26,22 +50,16 @@ def predict_text(image):
     num_chars = 5  # Estimated number of characters in the word
     segment_width = image.shape[1] // num_chars  # Split image into equal parts
 
-    predicted_text = []
-    for i in range(num_chars):
-        char_segment = image[:, i * segment_width:(i + 1) * segment_width, :]
-
-        # Resize each character segment to (64, 64, 3) to match model input
-        char_segment = Image.fromarray((char_segment * 255).astype(np.uint8))  # Convert back to PIL
-        char_segment = char_segment.resize((64, 64))  # Resize to match model input
-        char_segment = np.array(char_segment) / 255.0  # Normalize again
-        char_segment = np.expand_dims(char_segment, axis=0)  # Add batch dimension
-
-        # Predict character for each segment
-        prediction = model.predict(char_segment)
-        char = decode_prediction(prediction)
-        predicted_text.append(char)
-
-    return "".join(predicted_text)
+def predict_text(image):
+    processed_image = preprocess_image(image)  # Make sure the image is preprocessed to (64, 64, 3)
+    prediction = model.predict(processed_image)
+    
+    print("Model output shape:", prediction.shape)  # Should be (1, 78)
+    print("Model output values:", prediction)        # Check the raw probabilities
+    
+    # Decode the prediction to get the drug name
+    predicted_text = decode_prediction(prediction)
+    return predicted_text
 
 
 import numpy as np