How do I get the starting timestamp of a pipeline? Here is a part of my code starting the pipelines. I am trying to record and encode camera streams of two cameras at once and need the starting timestamp for recording/encoding for each of the cameras. ``` with contextlib.ExitStack() as stack: deviceInfos = dai.Device.getAllAvailableDevices() print("=== Found devices: ", deviceInfos) queues: dict = {} pipelines = [] # Create a pipeline and queues for each connected device for deviceInfo in deviceInfos: # Create pipeline context pipeline = stack.enter_context(dai.Pipeline()) device = pipeline.getDefaultDevice() device.setIrLaserDotProjectorIntensity(0) # Enhancement of depth perception device.setIrFloodLightIntensity(0) # Enhancement of low light performance mxId = get_device_info(device, deviceInfo) # Create and start pipeline for each device pipeline, output = self.createPipeline(pipeline, mxId) pipeline.start() camera_start[mxId] = datetime.now() logging.info(f"Camera [{camera_start[mxId]}] started recording at: {datetime.now()}") pipelines.append(pipeline) ``` I guess that there must be some things happening asynchronously in pipeline.start(), which is messing with the camera_start points.

Hey @"Jburoko"#3164 — `pipeline.start()` returns before sensors/encoders emit data, so `datetime.now()` there won’t match when recording actually begins. Use the timestamp on the first packet you actually record (e.g., `video`/`encOut`) via `getTimestamp()` (host-synced).

@"OskarSonc"#p31925 I just gave you access to it

Camera recording starting timestamp

Jburoko

Hey @OskarSonc
So I used the following example code for video encoding:
https://docs.luxonis.com/software-v3/depthai/examples/video_encoder/video_encode/
(But removed the part that displays the shows the video with cv2.imshow).

Where/How would I need to apply the getTimestamp function in that code? Could you maybe give me an example?

OskarSonc

You could put it into def process(self, frame):, we don't have an example for exactly this…
Did you check out our multiple-device setup and examples?
https://docs.luxonis.com/software-v3/depthai/tutorials/multi-device-setup

luxonis/oak-examplestree/main/tutorials/multiple-devices

Jburoko

OskarSonc

I actually tried your solution again. But still have a delay between the two camera streams 😥 .
Can you maybe tell me what could be the issue there? 🤔

Jburoko

I actually checked out the multiple-device setup in the link you sent me before asking the question.
I solved it now by creating all pipelines at first and later on starting the pipelines in a separate loop and immediately after storing the timestamps.
This solved my issue at least.

OskarSonc

Hey, I thought you resolved your issue before ...
Could you tell more and maybe share the MRE?

Jburoko

Yes you're right. I thought that issue has been solved.
But after multiple tests I noticed that there was still a delay.

So what I am trying to do is to record with two cameras simultaneously. Therefore, the camera pipelines are started consecutively and the timestamps after each camera processing start (VideoSaver class) is stored.
Then each time a light barrier is triggered a timestamp is saved as start or end for time windows.

After recording is finished - Recording will automatically stop after 10 sec in the MRE - the timewindows are cut out of both camera streams and I notice a delay between both cameras.

An MRE can be found here:
https://drive.google.com/drive/folders/1IWcf6tOcybVCRYp8gcW2tIGw_hmlqYDL?usp=drive_link

To start the script you need to run app/features/modules/camera.py
And I defined the camera-specs depending on their camera-id under app/camera_specs/*

I hope this helps

OskarSonc

@Jburoko - thanks for sharing this, but I don't have access to the drive files you sent ...

Jburoko

OskarSonc
I just gave you access to it

OskarSonc

Hey @Jburoko — I got your MRE running with both cameras. I noticed the two spec files set different FPS per camera; if that isn’t intentional, matching them will avoid steady frame-count drift. On startup I see the pipelines begin about 85 ms apart (e.g., …37.168180 vs …37.257924), which is expected since the script starts them sequentially and logs datetime.now() on the host. Despite that, the saved clips look synced on my side—I pointed both cams at a miliseconds stopwatch on-screen and the timestamps matched frame-for-frame (see images). Could you describe exactly where you’re seeing the offset—do the clipped videos drift, or is it a constant shift from how the windows are computed?
My idea is to switch the cut logic to use each frame’s getTimestamp() (host-synced)

Jburoko

Hi OskarSonc
The different FPS are intentional 🙂.
Actually, I am having a constant shift between the frames. Like a difference of 1-2sec.
I also tried 'depthai.Clock.now()' instead of 'datetime.now()', if that is what you meant. But I am still having the same issue.

FYI - Maybe it has an influence:
In the shared files I sent you, there is only the camera logic. But I am running the camera script concurrently with a shiny (Python package) frontend. Could it be possible that they interfere with each other? However, when using one camera, I didn't have any issues.

OskarSonc

Got it — a constant 1–2 s shift screams “cutting by host time / UI backlog.” Cut windows by each frame’s getTimestamp() (host-synced), set queues to maxSize=4, blocking=True, and re-run once without the shiny app to confirm it’s host contention.

Jburoko

OskarSonc

Ok. I will be able to try it on Friday again. Then I will give you an update.