Update app.py

app.py

@@ -15,54 +15,33 @@ from PIL import Image
 # LOAD THE MODEL
 # ------------------------------------
 
-# load the float32 TFLite model
 interpreter = Interpreter(model_path="resnet50_float32.tflite")
-
-# allocate memory for the model's input and output tensors
 interpreter.allocate_tensors()
 
-# get input and output tensor details
 input_details  = interpreter.get_input_details()
 output_details = interpreter.get_output_details()
 
-# image size ResNet50 expects
 INPUT_SIZE = (224, 224)
 
 print("Gatekeeper model loaded successfully")
 
 
 # ------------------------------------
-# THRESHOLDS
+# THRESHOLD
 # ------------------------------------
 
-# minimum probability for cervix to be accepted as a positive detection
+# cervix must score at least 0.55 to be accepted as a positive detection
 CERVIX_THRESHOLD = 0.55
 
-# minimum gap between cervix and non-cervix probabilities
-# if the gap is smaller than this the prediction is too uncertain to trust
-CONFIDENCE_GAP = 0.15
-
-# minimum image brightness - images below this are too dark to classify
-MIN_BRIGHTNESS = 30
-
-# minimum image contrast - images below this are blank or uniform
-MIN_STD = 20
-
 
 # ------------------------------------
 # IMAGE PREPROCESSING FUNCTION
 # ------------------------------------
 
 def preprocess_image(image):
-    # convert numpy array to PIL Image in RGB format and resize
     img = Image.fromarray(image).convert("RGB").resize(INPUT_SIZE)
-
-    # convert to float32 numpy array and normalise to [0, 1]
     img = np.array(img, dtype=np.float32) / 255.0
-
-    # add batch dimension: (224, 224, 3) → (1, 224, 224, 3)
     img = np.expand_dims(img, axis=0)
-
     return img
 
 
@@ -71,50 +50,27 @@ def preprocess_image(image):
 # ------------------------------------
 
 def classify_image(image):
-    # if the user submits without an image return a warning
     if image is None:
         return None, "Please upload an image first"
 
-    # Option 3: Basic image sanity checks
-    img_array = np.array(image)
-
-    # reject images that are too dark to analyse reliably
-    if img_array.mean() < MIN_BRIGHTNESS:
-        return None, "Image is too dark - please upload a clearer photo"
-
-    # reject images that are blank, uniformly coloured, or plain screenshots
-    if img_array.std() < MIN_STD:
-        return None, "Image appears blank or uniform - please upload a real photo"
-
-    # preprocess the image
+    # preprocess and run inference
     processed = preprocess_image(image)
-
-    # load the preprocessed image into the model's input tensor
     interpreter.set_tensor(input_details[0]['index'], processed)
-
-    # run inference
     interpreter.invoke()
-
-    # read the output tensor
     output = interpreter.get_tensor(output_details[0]['index'])
+
     print(f"Raw model output: {output}")
 
-    # extract individual class probabilities
     prob_non_cervix = float(output[0][0])
     prob_cervix     = float(output[0][1])
 
     print(f"Non-Cervix: {prob_non_cervix:.4f} | Cervix: {prob_cervix:.4f}")
 
-    # Option 1: Confidence threshold + gap check
-    gap = prob_cervix - prob_non_cervix
-
-    if prob_cervix >= CERVIX_THRESHOLD and gap >= CONFIDENCE_GAP:
+    # simple threshold check
+    if prob_cervix >= CERVIX_THRESHOLD:
         prediction_text = "Cervix Detected"
-    elif prob_non_cervix >= CERVIX_THRESHOLD and gap <= -CONFIDENCE_GAP:
-        prediction_text = "Non-Cervix"
     else:
-        # not confident enough either way - temporary misclassification safety net
-        prediction_text = "Uncertain - please retake or upload a clearer image"
+        prediction_text = "Non-Cervix"
 
     scores = {
         "Cervix":     round(prob_cervix, 4),
@@ -173,12 +129,6 @@ with gr.Blocks(theme=gr.themes.Soft()) as app:
     | 0     | Non-Cervix  | Image does NOT contain cervix    |
     | 1     | Cervix      | Image contains cervix            |
 
-    ---
-    **How predictions work:**
-    - **Cervix Detected** - model scored >= 0.55 with a gap of >= 0.15 over Non-Cervix
-    - **Non-Cervix** - model scored >= 0.55 with a gap of >= 0.15 over Cervix
-    - **Uncertain** - model was not confident enough; retake the image
-
     ---
     Disclaimer: This tool is for research purposes only.
     It is not intended for clinical diagnosis or medical use.