fix(axelera): never attribute primer/keep-alive results to a real request

The SDK can buffer primer results and emit them AFTER the first real frame
is already pending; the FIFO pump then popped a real id and stored the
skipped/buffered (empty) result under it, dropping the real one. This
misattributes detections across cameras even with run() serialized.

- eliminate keep-alive frames entirely (Frigate feeds the queue
  continuously; nothing to keep alive) - a synthetic frame's result can
  arrive during a pending real id,
- drain the primer results by index BEFORE any pending-id bookkeeping, so
  they can never consume a request id; the first run() waits up to 0.5s
  for the primer drain.

frigate/detectors/plugins/axelera.py

@@ -51,6 +51,7 @@ AX_DEFAULT_NETWORK = "yolox-s-coco-onnx"
 
 INFERENCE_TIMEOUT = 5.0
 MAX_ROWS = 20
+PRIMER_FRAMES = 3
 
 
 # ----------------- Utility Functions ----------------- #
@@ -152,6 +153,9 @@ class _AxeleraInference:
         # flight the FIFO pop can misattribute a result to the wrong camera.
         # With at most one frame in flight the result can only belong to it.
         self._run_lock = threading.Lock()
+        # Set once the SDK has finished emitting the primer results, so they
+        # can never be attributed to a real pending request id.
+        self._primed = threading.Event()
 
         if stream_factory is None:
             # lazy import: never import the SDK unless the detector is used
@@ -177,20 +181,19 @@ class _AxeleraInference:
 
     def _feed(self):
         # primer frames: the SDK init pre-buffers several frames before the
-        # source yields (validated on hardware). Their ids are NOT registered,
-        # so their stream results are dropped by the pump.
-        for _ in range(3):
+        # source yields (validated on hardware). Their results are drained by
+        # _pump (by index) and NEVER attributed to a request id.
+        for _ in range(PRIMER_FRAMES):
             yield np.zeros((self._height, self._width, 3), np.uint8)
         while True:
             try:
                 frame_id, frame = self._frame_q.get(timeout=1.0)
             except queue.Empty:
-                # keep-alive: no request pending, keep the SDK stream fed so the
-                # inference timeout never kills it during quiet periods. The
-                # None id means this frame's result is dropped by the pump.
-                frame_id, frame = None, np.zeros(
-                    (self._height, self._width, 3), np.uint8
-                )
+                # Never synthesize keep-alive frames: a synthetic frame's
+                # result can arrive while a real id is pending and steal that
+                # id (misattribution across cameras). In Frigate the queue is
+                # fed continuously, so idling simply waits here.
+                continue
             if frame is None:  # shutdown sentinel from stop()
                 return
             if frame_id is not None:
@@ -252,7 +255,14 @@ class _AxeleraInference:
     def _pump(self):
         logger.info("axelera: inference stream started")
         try:
-            for frame_result in self._stream:
+            for run_index, frame_result in enumerate(self._stream):
+                if run_index < PRIMER_FRAMES:
+                    # Primer results are drained WITHOUT consuming a request
+                    # id — otherwise a buffered primer result could steal a
+                    # real id that is already pending (P1 misattribution).
+                    if run_index == PRIMER_FRAMES - 1:
+                        self._primed.set()
+                    continue
                 out = _pad_rows(self._detection_rows(frame_result))
                 if self._pending_ids:
                     # FIFO: the stream returns results in submission order, so
@@ -299,6 +309,12 @@ class _AxeleraInference:
             # permanently wedging this detector.
             raise RuntimeError("Axelera inference thread stopped; restart required.")
 
+        # Wait briefly for the primer results to drain ONCE, so they can never
+        # be attributed to a real request id. After this the pump only ever
+        # sees real frames (keep-alive frames are no longer emitted).
+        if not self._primed.is_set():
+            self._primed.wait(timeout=0.5)
+
         # Every submitted frame gets a unique id; the pump stores the result
         # under this id, so we can wait for OUR frame's outcome exclusively.
         self._next_id += 1