How to use multiple cameras running the same model

Andrew

I have a custom model running on an Oak D Lite using the detection network example program, but I need to add an additional Oak D Lite to the setup. What do I need to do in the code to get a second camera also running the model?

OskarSonc

Hey @Andrew
You can check out this page:
https://docs.luxonis.com/software-v3/depthai/tutorials/multi-device-setup

Thanks,
Oskar

Andrew

@OskarSonc

That page was helpful and I was able to see output for both of my cameras. However I tried to alter the example code they have there to run my model, but I wasn't able to link a detection network to the outputs without getting "possible device error/misconfiguration." errors. I tried to run some of the demo scripts on the page, but I got the same errors.

Thanks, Andrew

Andrew

I've worked on my code a bit more and I only get the error messages when I try to run the code with both cameras plugged in. When I only have one plugged in I get no errors, but I see no output.

OskarSonc

Hey @Andrew,

That error usually means the pipeline or model node isn’t matching what the device expects — e.g. wrong NN node type, mismatched input size, or reusing the same pipeline for both cameras. Try:

Creating a separate pipeline per device (one dai.Pipeline() each).
Linking cam.preview → nn.input and making sure the input size matches your model.
Using the correct NN node (YoloDetectionNetwork, MobileNetDetectionNetwork, or NeuralNetwork) for your blob type.
Testing with both cameras on separate USB ports or a powered hub.
enabling DEPTHAI_LEVEL='debug' too see what is happening

If all that fails could you share a small code snippet (MRE) and your blob details? That’ll help us pinpoint the cause.

Andrew

I managed to get the code to work when only one camera is connected so I'm pretty sure I have the NN and camera nodes configured properly. I'm using YoloV11. Each camera has its own pipeline and right now I'm creating a NN node for each camera, but for whatever reason it only works for the first camera (when both are connected the first camera displays

. Here is the part of the code that seems to be the issue:

OskarSonc

Hey @Andrew — thanks for the snippet. I see you are using Raspberry Pi right? I think the issue is those implicit host queues:
detectionNetwork.passthrough.createOutputQueue()
detectionNetwork.out.createOutputQueue()
When two devices run in the same process, these auto-named streams collide, so cam #1 binds fine and cam #2 throws X_LINK / reconnect errors.

Use device-scoped outputs with unique stream IDs per device (avoid implicit createOutputQueue() on node outputs).
On Pi, plug cams into a powered USB 3 hub and a solid PSU (≥5V/3A).
Reduce bandwidth to test stability: NV12 preview at your NN input size, 10–15 FPS.
Stagger opens (100–300 ms between devices).

Andrew

OskarSonc I think I understand what you're saying, but after looking at the documentation I'm not sure how to make the createOutputQueue() calls explicit

OskarSonc

Hey @Andrew — to avoid stream collisions with two devices, build one Pipeline(device) per camera and call .createOutputQueue() on the node outputs from that same pipeline (e.g., requestOutput(...) or nn.out). Those queues are already device-scoped, so you don’t name them. This is exactly how we do it here:
luxonis/oak-examplesblob/main/tutorials/multiple-devices/multiple-devices-preview/multi-device-yolov6.py

If issues persist, enable DEBUG logs to confirm each stream binds to the right MXID.

Andrew

OskarSonc I altered my code to better emulate the example you included, but I got the same issue. I tried just running that example code without any modifications, but I still had the same errors when I connect multiple cameras.

OskarSonc

@Andrew thanks for info.
So I tried running the same example as you(multi-device-yolov6.py) and it worked without any problems on my PC.
That being said, I think the code is not the problem in your case... I think it could be Raspberry Pi’s USB power/bandwidth related...
Is there any way you could test this on a normal PC or maybe try using a powered USB 3.0 hub/power adapter?
If I understand correctly, this example worked:
luxonis/oak-examplesblob/main/tutorials/multiple-devices/multiple-devices-preview/main.py

Andrew

I just tried running it on another PC and it seemed to work. I ordered some of the usb-C y splitters so hopefully that will fix the issue with power delivery.
Thank you for your help!

My RPI only has one set of USB 3.0 ports, if I'm using the splitters to get power delivery from another source will there still be a data bandwidth problem if both cameras are connected to the same usb hub?

OskarSonc

Hey @Andrew — this looks mostly like power. The splitters should stabilize things.
If it turns out to be bandwidth too, we can tweak the pipeline (NV12 at NN input size, lower FPS, or use H.265/MJPEG for high-res) to cut bus load. Keep us posted!