Update app.py

app.py

@@ -1,27 +1,30 @@
 import gradio as gr
 import tensorflow as tf
 import numpy as np
-from PIL import Image
 import cv2
 
-# Load model
+# Load Teachable Machine model
 model = tf.keras.models.load_model("keras_model.h5")
 
-# Replace this with your actual label list if you have it
+# Labels from your model — change if needed
 labels = ["Good Posture", "Bad Posture"]
 
 def predict_from_webcam(frame):
-    # Resize frame to match model input size (usually 224x224 for Teachable Machine)
+    # Convert to RGB if needed (Gradio usually provides RGB)
+    if frame.shape[-1] == 4:
+        frame = frame[:, :, :3]  # remove alpha channel
+
+    # Resize to 224x224 (Teachable Machine default)
     img = cv2.resize(frame, (224, 224))
-    img = img / 255.0  # normalize
-    img = np.expand_dims(img, axis=0)  # batch dimension
+    img = img / 255.0  # normalize to [0, 1]
+    img = np.expand_dims(img, axis=0)  # add batch dimension
 
-    # Predict
+    # Prediction
     prediction = model.predict(img)[0]
-    top_label = labels[np.argmax(prediction)]
+    label = labels[np.argmax(prediction)]
     confidence = np.max(prediction)
 
-    return f"{top_label} ({confidence*100:.2f}%)"
+    return f"{label} ({confidence * 100:.2f}%)"
 
 # Gradio interface
 demo = gr.Interface(
@@ -29,8 +32,8 @@ demo = gr.Interface(
     inputs=gr.Image(source="webcam", streaming=True),
     outputs=gr.Text(),
     live=True,
-    title="Posture Detector",
-    description="Detects if your posture is good or bad using a webcam model trained with Teachable Machine."
+    title="🪑 Posture Detector",
+    description="Detects good or bad posture from webcam using a Teachable Machine model."
 )
 
 demo.launch()