The idle heartbeat check in BirdsEyeOutputProcess.update() compares
time.monotonic() (seconds since an arbitrary point, typically boot)
against last_output_time which is set from datetime.datetime.now().timestamp()
(Unix epoch seconds).
These are completely different time bases. The subtraction produces a
large negative number, so the idle heartbeat condition can never be
satisfied. This means birdseye stops sending frames when all cameras
go idle, instead of continuing at the configured idle_heartbeat_fps.
Use datetime.datetime.now().timestamp() consistently for both the
heartbeat check and the output time tracking.
* fix: operator precedence bug in detection type check
The condition:
topic == DetectionTypeEnum.api.value or DetectionTypeEnum.lpr.value
evaluates as:
(topic == DetectionTypeEnum.api.value) or (DetectionTypeEnum.lpr.value)
Since DetectionTypeEnum.lpr.value is a non-empty string (truthy), the
second operand is always True regardless of topic. The intended check
is whether topic matches either enum value:
topic == DetectionTypeEnum.api.value or topic == DetectionTypeEnum.lpr.value
* fix: apply same or operator fix to review/maintainer.py
Same issue as record/maintainer.py — the condition was always true
because the bare enum value is truthy.
* style: ruff format record/maintainer.py
setVolumeStates was replacing the entire state object instead of
merging, so changing one camera's volume reset all others to default.
Uses the functional update pattern to preserve existing state, matching
how toggleAudio already works.
* fix: check HTTP response status before parsing JSON body
upload_image() calls r.json() before checking r.ok. If the server
returns an error response (401, 500, etc) with a non-JSON body,
this raises a confusing JSONDecodeError instead of the intended
'Unable to get signed urls' error message.
Move the r.ok check before the r.json() call.
* style: remove extra blank line for ruff
The name parameter was interpolated directly into the SQL query via
f-string, allowing SQL injection through crafted face name values.
Use a parameterized query with ? placeholder instead.
parse_preset_input() uses input[len(_user_agent_args) + 1] to find
the FPS placeholder, but preset-http-jpeg-generic does not include
_user_agent_args at the start of its list (only preset-http-mjpeg-generic
does). The FPS placeholder '{}' is at index 1, not index 3.
This means the detect_fps value overwrites '-1' (the stream_loop
argument) instead of the '{}' FPS placeholder, so the preset always
uses the literal string '{}' as the framerate.
When hwaccel_args is a list (not a preset string), the fallback in
parse_preset_hardware_acceleration_encode() calls
arg_map["default"].format(input, output) with only 2 positional args.
But PRESETS_HW_ACCEL_ENCODE_BIRDSEYE["default"] contains {0}, {1}, {2}
expecting ffmpeg_path as the first arg.
This causes IndexError: Replacement index 2 out of range for size 2
which crashes create_config.py on every go2rtc start, taking down
all camera streams.
Pass ffmpeg_path as the first argument to match the preset template.
In BirdsEyeFrameManager.__init__(), the numpy slice that copies the
custom logo (transparent_layer from custom.png alpha channel) onto
blank_frame has shape[0] and shape[1] swapped:
blank_frame[y:y+shape[1], x:x+shape[0]] = transparent_layer
shape[0] is rows (height) and shape[1] is cols (width), so the row
range needs shape[0] and the column range needs shape[1]:
blank_frame[y:y+shape[0], x:x+shape[1]] = transparent_layer
The bug is masked for square images where shape[0]==shape[1]. For
non-square images (e.g. 1920x1080), it produces:
ValueError: could not broadcast input array from shape (1080,1920)
into shape (1620,1080)
This silently kills the birdseye output process -- no frames are
written to the FIFO pipe, go2rtc exec ffmpeg times out, and the
birdseye restream shows a black screen with no errors in the UI.
In both expire_snapshots() and expire_clips(), the expired_events
query uses .iterator() for lazy evaluation, but the very next line
calls list(expired_events) inside an f-string for debug logging.
This consumes the entire iterator, so the subsequent for loop that
deletes media files from disk iterates over an exhausted iterator
and processes zero events.
Snapshots and clips for removed cameras are never deleted from disk,
causing gradual disk space exhaustion.
Materialize the iterator into a list before logging so both the
debug message and the cleanup loop use the same data.
The cosine similarity calculation is guarded by:
if not np.any(np.linalg.norm(velocities, axis=1))
This enters the block when ALL velocity norms are zero, then divides
by those zero norms. The condition should check that all norms are
non-zero before computing cosine similarity:
if np.all(np.linalg.norm(velocities, axis=1))
Also fixes debug log that shows average_velocity[0] for both x and y
velocity components (second should be average_velocity[1]).
The connect() function creates a WebSocket but never stores the
reference. The useEffect cleanup only closes the RTCPeerConnection
via pcRef, leaving the WebSocket open.
Each time the component re-renders with changed deps (camera switch,
playback toggle, microphone toggle), a new WebSocket is created
without closing the previous one. This leaks connections until the
browser garbage-collects them or the server times out.
Store the WebSocket in a ref and close it in the cleanup function.
cv2.imread with IMREAD_UNCHANGED loads the image as-is, but the code
unconditionally indexes channel 3 (birdseye_logo[:, :, 3]) assuming
RGBA format. This crashes with IndexError for:
- Grayscale PNGs (2D array, no channel dimension)
- RGB PNGs without alpha (3 channels, no index 3)
- Fully transparent PNGs saved as grayscale+alpha (2 channels)
Handle all image formats:
- 2D (grayscale): use directly as luminance
- 4+ channels (RGBA): extract alpha channel (existing behavior)
- 3 channels (RGB/BGR): convert to grayscale
Also fixes the shape[0]/shape[1] swap in the array slice that breaks
non-square images (related to #6802, #7863).
In BirdsEyeFrameManager.update(), the exception handler on line 756
resets self.active_cameras to [] (a list), but it is initialized as
set() and compared as a set throughout the rest of the code.
Since set() \!= [] evaluates to True even though both are empty, the
next call to update_frame() will incorrectly detect a layout change
and trigger an unnecessary frame rebuild after every exception.
escape_special_characters() returns a ValueError object instead of
raising it when the input path exceeds 1000 characters. The exception
object gets used as a string downstream instead of triggering error
handling.
When an existing tracked object's label or stationary status changes
(e.g. sub_label assignment from face recognition), the update handler
declares a new const newObjects that shadows the outer let newObjects.
The label and stationary mutations apply to the inner copy, but
handleSetObjects on line 148 reads the outer variable which was never
mutated. The update is silently discarded.
Remove the inner declaration so mutations apply to the outer variable
that gets passed to handleSetObjects.
gpu <= len(self._valid_gpus) should be gpu < len(self._valid_gpus).
The list is zero-indexed, so requesting gpu index equal to the list
length causes an IndexError. For example, with 2 valid GPUs (indices
0 and 1), requesting gpu=2 passes the check (2 <= 2) but
self._valid_gpus[2] is out of bounds.
