From 71ccf55927a283b6d44956d5f8a7a7d291cefc6c Mon Sep 17 00:00:00 2001
From: patrykorwat <patrykorwat@gmail.com>
Date: Thu, 20 Aug 2026 19:49:05 +0000
Subject: [PATCH] fix(axelera): serialize run() to prevent cross-camera result
 misattribution

Voyager SDK does not tag results with a per-frame id and can reorder or
buffer stream results under sustained load (observed on live with many
thousands of frames: camera A started receiving boxes/paths from camera B).
With several frames in flight, the FIFO pairing can store a result under a
different frame's id.

run() now holds a lock so at most one frame is in flight; the next stream
result can only belong to the frame just pushed, making pairing exact
regardless of SDK reordering. Pipeline throughput stays far below limits
(~9ms per inference vs 28ms available per frame at 7x5fps).
---
 frigate/detectors/plugins/axelera.py | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/frigate/detectors/plugins/axelera.py b/frigate/detectors/plugins/axelera.py
index 1d1153f9ed..b25a4eb50b 100644
--- a/frigate/detectors/plugins/axelera.py
+++ b/frigate/detectors/plugins/axelera.py
@@ -146,6 +146,12 @@ class _AxeleraInference:
         self._pending_ids: collections.deque = collections.deque()
         self._results: dict[int, np.ndarray] = {}
         self._next_id = 0
+        # Serializes run() calls: the SDK stream does not tag results with a
+        # per-frame id and can reorder/buffer results under sustained load
+        # (observed after many thousands of frames), so with several frames in
+        # 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()
 
         if stream_factory is None:
             # lazy import: never import the SDK unless the detector is used
@@ -262,6 +268,14 @@ class _AxeleraInference:
 
     def run(self, tensor_input, timeout: float = INFERENCE_TIMEOUT) -> np.ndarray:
         """Run one inference and return this frame's detections (see detect_raw)."""
+        # At most ONE frame may be in flight at a time: with several frames in
+        # the stream the SDK can reorder/buffer their results (misattribution
+        # across cameras observed after many thousands of frames). Serialized,
+        # the next returned result can only belong to the one frame we pushed.
+        with self._run_lock:
+            return self._run_locked(tensor_input, timeout)
+
+    def _run_locked(self, tensor_input, timeout: float) -> np.ndarray:
         if tensor_input is None or getattr(tensor_input, "ndim", 0) < 3:
             raise ValueError(
                 f"Unexpected frame shape: {getattr(tensor_input, 'shape', tensor_input)}"
@@ -316,7 +330,7 @@ class _AxeleraInference:
     def stop(self):
         # cleanup must not raise
         try:
-            self._frame_q.put_nowait(None)
+            self._frame_q.put_nowait((None, None))
         except Exception:  # noqa: S110, BLE001
             pass
         try: