Update app.py

app.py

@@ -253,33 +253,24 @@ def run(frame, sequence_state):
 
     return out_rgb, probs_dict, sequence_state
 
-# Legacy Gradio-compatible state
-sequence_state = []
-
-def predict(frame):
-    global sequence_state
-
-    if frame is None:
-        return None, {"(no frame)": 1.0}
-
-    # Sometimes legacy Gradio provides PIL
-    if isinstance(frame, Image.Image):
-        frame = np.array(frame)
-
-    out_img, probs_dict, sequence_state = run(frame, sequence_state)
-    return out_img, probs_dict
-
-demo = gr.Interface(
-    fn=predict,
-    inputs=gr.inputs.Image(source="webcam", type="numpy", label="Webcam"),
-    outputs=[
-        gr.outputs.Image(type="numpy", label="Output (Annotated)"),
-        gr.outputs.Label(num_top_classes=5, label="Probabilities (Top 5)"),
-    ],
-    title="Live Sign Language Gesture Demo (CNN-LSTM + Multi-Head Attention)",
-    description="Show your hand gesture to the webcam. Prediction starts after 30 frames are collected.",
-    live=True,
-)
+with gr.Blocks() as demo:
+    gr.Markdown("# Live Sign Language Gesture Demo (CNN-LSTM + Multi-Head Attention)")
+    gr.Markdown("Show your hand gesture to the webcam. Prediction starts after 30 frames are collected.")
+
+    seq_state = gr.State([])
+
+    with gr.Row():
+        cam = gr.Image(sources=["webcam"], type="numpy", label="Webcam")
+        out_img = gr.Image(type="numpy", label="Output (Annotated)")
+
+    out_label = gr.Label(num_top_classes=5, label="Probabilities (Top 5)")
+
+    # Stream events are supported in gradio 4.x
+    cam.stream(
+        fn=run,
+        inputs=[cam, seq_state],
+        outputs=[out_img, out_label, seq_state],
+    )
 
 if __name__ == "__main__":
     demo.launch()