Merge branch 'main' into rf-detr-exp-adapter-fix

Author: hansent

inference/core/interfaces/webrtc_worker/webrtc.py

@@ -13,7 +13,7 @@
     RTCSessionDescription,
     VideoStreamTrack,
 )
-from aiortc.contrib.media import MediaPlayer, MediaRelay
+from aiortc.contrib.media import MediaPlayer, MediaRelay, PlayerStreamTrack
 from aiortc.rtcrtpreceiver import RemoteStreamTrack
 from av import VideoFrame
 from av import logging as av_logging
@@ -130,6 +130,9 @@ async def recv(self):
             logger.info("Timeout reached, terminating inference pipeline")
             self._terminate_event.set()
 
+        if isinstance(self.track, PlayerStreamTrack):
+            if self.track._queue.qsize() > 30:
+                self.track._queue.get_nowait()
         frame: VideoFrame = await self.track.recv()
 
         self._received_frames += 1
@@ -317,17 +320,15 @@ async def init_rtc_peer_connection_with_loop(
         logger.info("Processing RTSP URL: %s", webrtc_request.rtsp_url)
         player = MediaPlayer(
             webrtc_request.rtsp_url,
+            format="rtsp",
             options={
                 "rtsp_transport": "tcp",
                 "rtsp_flags": "prefer_tcp",
                 "stimeout": "2000000",  # 2s socket timeout
             },
         )
         video_transform_track.set_track(
-            track=relay.subscribe(
-                player.video,
-                buffered=False if webrtc_request.webrtc_realtime_processing else True,
-            )
+            track=player.video,
         )
         peer_connection.addTrack(video_transform_track)
 

inference/core/version.py

@@ -1,4 +1,4 @@
-__version__ = "0.59.2"
+__version__ = "0.60.0"
 
 
 if __name__ == "__main__":

inference/models/perception_encoder/vision_encoder/tokenizer.py

@@ -45,14 +45,17 @@ def bytes_to_unicode():
         + list(range(ord("¡"), ord("¬") + 1))
         + list(range(ord("®"), ord("ÿ") + 1))
     )
+    # Precompute set for faster lookup, eliminate repeated containment checks
+    bs_set = set(bs)
     cs = bs[:]
     n = 0
-    for b in range(2**8):
-        if b not in bs:
+    for b in range(256):  # 2**8
+        if b not in bs_set:
             bs.append(b)
-            cs.append(2**8 + n)
+            cs.append(256 + n)
             n += 1
-    cs = [chr(n) for n in cs]
+    # Use list comprehension and map for efficient conversion
+    cs = list(map(chr, cs))
     return dict(zip(bs, cs))