Update app.py

app.py

@@ -1,6 +1,6 @@
 import gradio as gr
 import numpy as np
-from PIL import Image
+from PIL import Image, ImageFilter, ImageStat
 import tensorflow as tf
 
 # Load TFLite model
@@ -13,13 +13,39 @@ output_details = interpreter.get_output_details()
 
 # Class labels
 class_names = ['Dyskeratotic', 'Koilocytotic', 'Metaplastic', 'Parabasal', 'Superficial-Intermediat']
-CONFIDENCE_THRESHOLD = 0.7
+CONFIDENCE_THRESHOLD = 0.5
+
+def is_valid_cervical_image(image):
+    """Basic smart validation: checks variance, edges, brightness"""
+    # Convert to grayscale
+    gray = image.convert("L")
+    stat = ImageStat.Stat(gray)
+    
+    # Variance check (texture)
+    variance = stat.var[0]
+    if variance < 500:  # threshold, tweak as needed
+        return False, "Image lacks texture. Upload a proper cervical cell image."
+
+    # Edge detection check
+    edges = gray.filter(ImageFilter.FIND_EDGES)
+    edge_data = np.array(edges)
+    edge_pixels = np.sum(edge_data > 50)
+    if edge_pixels < 1000:  # too few edges
+        return False, "Image has too few edges. Upload a clear cervical cell image."
+
+    # Brightness check
+    brightness = stat.mean[0]
+    if brightness < 30 or brightness > 220:
+        return False, "Image brightness/contrast is not suitable. Adjust and upload again."
+
+    return True, ""
 
 def predict_image(image):
     try:
         # Validate input
-        if image.mode != "RGB":
-            return {"Error": 1.0}, "⚠️ Please upload a valid cervical cell image (RGB image required)."
+        valid, message = is_valid_cervical_image(image)
+        if not valid:
+            return {"Error": 1.0}, f"⚠️ {message}"
 
         # Preprocess
         image = image.resize((224, 224)).convert("RGB")
@@ -29,14 +55,14 @@ def predict_image(image):
         # Run inference
         interpreter.set_tensor(input_details[0]['index'], img_array)
         interpreter.invoke()
-        output = interpreter.get_tensor(output_details[0]['index'])[0]  # shape (num_classes,)
+        output = interpreter.get_tensor(output_details[0]['index'])[0]
 
         # Normalize
         probs = tf.nn.softmax(output).numpy()
 
-        # Check if prediction is below confidence threshold
+        # Confidence check
         if np.max(probs) < CONFIDENCE_THRESHOLD:
-            return {"Error": 1.0}, "⚠️ The model is unsure. Please upload a clearer/correct image of cervical cells."
+            return {"Error": 1.0}, "⚠️ The model is unsure. Please upload a clearer cervical cell image."
 
         # Convert to dict for Gradio Label
         probs_dict = {class_names[i]: float(probs[i]) for i in range(len(class_names))}
@@ -51,5 +77,5 @@ gr.Interface(
     inputs=gr.Image(type="pil"),
     outputs=[gr.Label(num_top_classes=len(class_names)), gr.Textbox()],
     title="Cervical Cancer Classification",
-    description="Upload an image. The model shows probabilities for each class and warns if the image is incorrect."
+    description="Upload a cervical cell image. The model shows probabilities for each class and warns if the image is incorrect."
 ).launch()