Update app.py

app.py

@@ -5,6 +5,7 @@ import gradio as gr
 import torch
 import torch.nn as nn
 import mediapipe as mp
+from PIL import Image
 
 # ----------------------------
 # Load labels (labels.json)
@@ -252,23 +253,33 @@ def run(frame, sequence_state):
 
     return out_rgb, probs_dict, sequence_state
 
-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.")
+# Legacy Gradio-compatible state
+sequence_state = []
 
-    seq_state = gr.State([])
+def predict(frame):
+    global sequence_state
 
-    with gr.Row():
-        cam = gr.Webcam(streaming=True, label="Webcam")
-        out_img = gr.Image(type="numpy", label="Output (Annotated)")
+    if frame is None:
+        return None, {"(no frame)": 1.0}
 
-    out_label = gr.Label(num_top_classes=5, label="Probabilities (Top 5)")
+    # Sometimes legacy Gradio provides PIL
+    if isinstance(frame, Image.Image):
+        frame = np.array(frame)
 
-    cam.stream(
-        fn=run,
-        inputs=[cam, seq_state],
-        outputs=[out_img, out_label, seq_state],
-    )
+    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,
+)
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()