Update pages/Camera.py

pages/Camera.py

@@ -1,37 +1,52 @@
 import logging
 import queue
 from collections import deque
-import asyncio
+import json
+import tempfile
 
 import streamlit as st
-from streamlit_webrtc import WebRtcMode, webrtc_streamer
-import av
+from streamlit_webrtc import WebRtcMode, webrtc_streamer, RTCConfiguration
+
 from utils import SLInference
 
 
 logger = logging.getLogger(__name__)
 
-def video_frame_callback(frame: av.VideoFrame, inference_thread, gestures_deque):
-    img = frame.to_ndarray(format="rgb24")
-
-    inference_thread.input_queue.append(img)
-    gesture = inference_thread.pred
-
-    if gesture not in ['no', '']:
-        if not gestures_deque:
-            gestures_deque.append(gesture)
-        elif gesture != gestures_deque[-1]:
-            gestures_deque.append(gesture)
-
-    return av.VideoFrame.from_ndarray(img, format="rgb24")
+RTC_CONFIGURATION = RTCConfiguration({
+    "iceServers": [
+        {"urls": ["stun:stun.l.google.com:19302"]},
+        {"urls": ["turn:TURN_SERVER_URL"], "username": "USERNAME", "credential": "CREDENTIAL"}
+    ]
+})
 
-def main(config_path):
+def main():
     """
     Main function of the app.
     """
-    inference_thread = SLInference(config_path)
+    config = {
+        "path_to_model": "S3D.onnx",
+        "threshold": 0.3,
+        "topk": 5,
+        "path_to_class_list": "RSL_class_list.txt",
+        "window_size": 32,
+        "provider": "OpenVINOExecutionProvider"
+    }
+
+    # Сохранение конфигурации во временный файл
+    with tempfile.NamedTemporaryFile(delete=False, mode='w', suffix='.json') as config_file:
+        json.dump(config, config_file)
+        config_file_path = config_file.name
+
+    inference_thread = SLInference(config_file_path)
     inference_thread.start()
 
+    webrtc_ctx = webrtc_streamer(
+        key="video-sendonly",
+        mode=WebRtcMode.SENDONLY,
+        rtc_configuration=RTC_CONFIGURATION,
+        media_stream_constraints={"video": True, "audio": False},
+    )
+
     gestures_deque = deque(maxlen=5)
 
     # Set up Streamlit interface
@@ -43,34 +58,37 @@ def main(config_path):
         """
         This application is designed to recognize sign language using a webcam feed.
         The model has been trained to recognize various sign language gestures and display the corresponding text in real-time.
-        
+
+        This demo app is based on code here: https://github.com/ai-forever/easy_sign
         The project is open for collaboration. If you have any suggestions or want to contribute, please feel free to reach out.
         """
     )
 
-    webrtc_ctx = webrtc_streamer(
-        key="sign-language-recognition",
-        mode=WebRtcMode.SENDRECV,
-        video_frame_callback=lambda frame: video_frame_callback(frame, inference_thread, gestures_deque),
-        media_stream_constraints={"video": True, "audio": False},
-        async_processing=True,
-    )
+    while True:
+        if webrtc_ctx.video_receiver:
+            try:
+                video_frame = webrtc_ctx.video_receiver.get_frame(timeout=1)
+            except queue.Empty:
+                logger.warning("Queue is empty")
+                continue
 
-    while webrtc_ctx.state.playing:
-        gesture = inference_thread.pred
+            img_rgb = video_frame.to_ndarray(format="rgb24")
+            image_place.image(img_rgb)
+            inference_thread.input_queue.append(video_frame.reformat(224, 224).to_ndarray(format="rgb24"))
 
-        if gesture not in ['no', '']:
-            if not gestures_deque:
-                gestures_deque.append(gesture)
-            elif gesture != gestures_deque[-1]:
-                gestures_deque.append(gesture)
+            gesture = inference_thread.pred
+            if gesture not in ['no', '']:
+                if not gestures_deque:
+                    gestures_deque.append(gesture)
+                elif gesture != gestures_deque[-1]:
+                    gestures_deque.append(gesture)
 
-        text_output.markdown(f'<p style="font-size:20px"> Current gesture: {gesture}</p>',
-                             unsafe_allow_html=True)
-        last_5_gestures.markdown(f'<p style="font-size:20px"> Last 5 gestures: {" ".join(gestures_deque)}</p>',
+            text_output.markdown(f'<p style="font-size:20px"> Current gesture: {gesture}</p>',
                                  unsafe_allow_html=True)
-        print(gestures_deque)
+            last_5_gestures.markdown(f'<p style="font-size:20px"> Last 5 gestures: {" ".join(gestures_deque)}</p>',
+                                 unsafe_allow_html=True)
+            print(gestures_deque)
 
 if __name__ == "__main__":
-    asyncio.set_event_loop(asyncio.new_event_loop())
-    main("configs/config.json")
+    main()
+