DepthAI can’t connect to an OAK camera

For USB OAK cameras, DepthAI can throw an error code like X_LINK_COMMUNICATION_NOT_OPEN or X_LINK_ERROR, which is usually a sign of a bad USB3 cable (or a USB2 cable). If you are using USB2 cable (and want USB2 bandwidth), you have to specify USB2 protocol, see Forcing USB2 Communication for more information. Another common issue is that users haven’t set udev rules on their Linux machine.

If you still can’t connect to the OAK camera, you should execute lsusb | grep 03e7. You should see a similar line:

$ lsusb | grep 03e7
Bus 001 Device 120: ID 03e7:2485 Intel Movidius MyriadX

Another thing to check is the dmesg -w. After executing that and pressing enter a few times (for separator), connect your OAK camera to the host. You should see a similar output in the terminal:

/~$ dmesg -w

[223940.862544] usb 1-3.2: new high-speed USB device number 120 using xhci_hcd
[223940.963357] usb 1-3.2: New USB device found, idVendor=03e7, idProduct=2485, bcdDevice= 0.01
[223940.963364] usb 1-3.2: New USB device strings: Mfr=1, Product=2, SerialNumber=3
[223940.963368] usb 1-3.2: Product: Movidius MyriadX
[223940.963371] usb 1-3.2: Manufacturer: Movidius Ltd.
[223940.963373] usb 1-3.2: SerialNumber: 03e72485

For PoE OAK cameras, see PoE Troubleshooting page.

If these commands didn’t return the expected log, see Support page.

Reporting firmware crash dump

Steps to create a crash dump:

  1. Run a code that causes the firmware crash

  2. Run Crash report script

Afterwards, please send the generated .txt file to us (email/github/forum), and our development team will try to fix the cause of the crash as soon as possible.

Ping was missed, closing the device connection

[host]   [warning] Monitor thread (device: 194435108198757340 []) - ping was missed, closing the device connection

This error is mostly seen on POE cameras, and can sometimes occur after a few hours or even days of running the pipeline. It happens because device doesn’t reply to ping messages, which usually happens if device is too busy (high CPU consumption), or there’s a networking issue. Potential solutions include:

  1. Lower OAK CPU consumption

If CPU usage is above 95% that can be a bad sign. You can measure CPU usage by running the depthai in debug mode. A few options to reduce CPU consumption are:

  • lower 3A FPS

  • Update networking settings on host computer - using ethtool, some have reported that increasing rx-usecs from 0 to 400 decreased Leon CPU usage from 99% to 93%

  • Reducing the pipeline / resolution / FPS, so there’s less computation happening on the camera

  1. Increasing watchdog timeout

Another solution is to increase watchdog timeout, so it isn’t as “trigger happy”. Default value for POE devices is 4000 ms, and user can set it to 4500ms (max value). This can be done by setting environmental variable DEPTHAI_WATCHDOG, so for example on linux: DEPTHAI_WATCHDOG=4500 python3

  1. Implementing re-connection

This would be quite recommended, as it would allow the application to recover from the error. So whenever there’s an error, re-connect to the device and continue running the pipeline.

For depthai API, this could be implemented as:

pipeline = dai.Pipeline()
# ...
while True:
  # Every time it crashes, re-initialize the device and upload the pipeline to it
  with dai.Device(pipeline) as device:
    queue = device.getOutputQueue("name")
    while True:
  1. Standalone mode

If you run your device in Standalone mode and communicate with it via networking protocols (TCP/UDP/HTTP/MQTT) and not XLink, there’s no ping mechanism, so this error won’t occur. If there were an issue on device side (which is why we have watchdog in the first place), device would auto-restart.

ImportError: No module named ‘depthai’

This indicates that the depthai was not found by your python interpreter. There are a handful of reasons this can fail:

  1. Is the Python API installed? Verify that it appears when you type:

    python3 -m pip list | grep depthai
  2. Are you using a supported platform for your operating system? If not, you can always install from source:

    cat /etc/os-release

Why is the Camera Calibration running slow?

Poor photo conditions can dramatically impact the image processing time) during the camera calibration. Under normal conditions, it should take 1 second or less to find the chessboard corners per-image on an Raspberry Pi but this exceed 20 seconds per-image in poor conditions. Tips on setting up proper photo conditions:

  • Ensure the checkerboard is not warped and is truly a flat surface. A high-quality option: print the checkerboard on a foam board.

  • Reduce glare on the checkerboard (for example, ensure there are no light sources close to the board like a desk lamp).

  • Reduce the amount of motion blur by trying to hold the checkerboard as still as possible.

Permission denied error

If python3 -m pip install fails with a Permission denied error, your user likely doesn’t have permission to install packages in the system-wide path.

[Errno 13] Permission denied: '/usr/local/lib/python3.7/dist-packages/...'

Try installing in your user’s home directory instead by adding the --user option. For example:

python3 -m pip install depthai --user

More information on Stackoverflow.

DepthAI does not show up under /dev/video* like web cameras do. Why?

The USB device enumeration could be checked with lsusb | grep 03e7 . It should print:

  • 03e7:2485 after reset (boot loader running)

  • 03e7:f63b after the application was loaded

No /dev/video* nodes are created. See OAK as a webcam if you would like to use OAK camera as a webcam.

Intermittent Connectivity with Long (2 meter) USB3 Cables

We’ve found that some hosts have trouble with long USB3 cables (above 6ft/2m). It seems to have something to do with the USB controller on the host side. For example, all Apple computers we’ve tested with have never exhibited the problem, as Apple computers have powerful USB controllers.

So, if you experience this problem with your host, there are potentially 3 options:

  1. Switching to a shorter USB3 cable (say 1 meter) will very likely make the problem disappear. These 1 meter (3.3 ft.) cables are a nice length and are shipped with OAK USB3 variants.

  2. Force USB2 mode. This will allow use of the long cable still, and many DepthAI use cases do not necessitate USB3 communication bandwidth - USB2 is plenty.

  3. Use Active USB3 cable. We have tested this 10m active cable and USB3 works as expected (even without powering the repeater).

Note that Ubuntu 16.04 has an independent USB3 issue, seemingly only on new machines though. We think this has to do w/ Ubuntu 16.04 being EOLed prior to or around when these new machines hit the market. For example, this computer (here) has rampant USB3 disconnect issues under Ubuntu 16.04 (with a 1 meter cable), but has none under Ubuntu 18.04 (with a 1 meter cable).

Forcing USB2 Communication

If you aren’t using a (working) USB3 cable or your host computer doesn’t support USB3, you should force the USB2 communication. It’s also recommended to use USB2 communication if you are using a longer USB cable (2m+).

For API usage, set the maxUsbSpeed=dai.UsbSpeed.HIGH when creating the dai.Device object

# Force USB2 communication
with dai.Device(pipeline, maxUsbSpeed=dai.UsbSpeed.HIGH) as device:
  # ...

If you are using depthai_demo you can specify USB speed with -usbs argument:

python3 -usbs usb2

Output from DepthAI keeps freezing

If the output from the device keeps freezing every few seconds, there may be a problem with the USB3 connection and forcing the device into USB2 mode could resolve this issue - instructions are in the chapter above.

When connection speed is USB2 (due to some hosts - Windows in particular - or USB controller/port/cable being USB2) - initialization of USB3-enabled firmware or streaming after a few frames may fail. The workaround here is to force the device to use the USB2-only firmware (mentioned in the chapter above).

DepthAI freezes after a few frames

If your app freezes and you don’t get any new messages from the device after a few messages (eg. 4 frames) after booting, it’s likely that queues filled up and were set to blocking mode. Additional details on node queues (on the OAK device) can be found here. We also recommend using Pipeline Graph tool to quikcly check if there is something wrong with the pipeline (eg. a node that isn’t connected to anything).

Udev rules on Linux

  • Failed to boot the device: 1.3-ma2480, err code 3

  • Failed to find device (ma2480), error message: X_LINK_DEVICE_NOT_FOUND

  • [warning] skipping X_LINK_UNBOOTED device having name "<error>"

  • Insufficient permissions to communicate with X_LINK_UNBOOTED device with name "1.1". Make sure udev rules are set

If you are getting any of the errors above, it’s most likely that udev rules are not set on your Linux machine.

To fix this, set the udev rules using the commands below, unplugging DepthAI and then plugging it back into USB afterwards.

echo 'SUBSYSTEM=="usb", ATTRS{idVendor}=="03e7", MODE="0666"' | sudo tee /etc/udev/rules.d/80-movidius.rules
sudo udevadm control --reload-rules && sudo udevadm trigger

CTRL-C Is Not Stopping It!

If you are trying to kill a program with CTLR-C, and it’s not working, try CTRL-\ instead. Usually this will work.

Nothing happening when running a DepthAI script

If upon running a DepthAI script, nothing happens (the script just hangs) and you cannot CTRL-C out of it. Try running script from the depthai-python/utilities folder. The script should error out with something like:

QObject::moveToThread: Current thread (0x55b9d0a00320) is not the object's thread (0x55b9d0cad7e0).
Cannot move to target thread (0x55b9d0a00320)


Aborted (Signal sent by tkill() 32104 1001)
Aborted (core dumped)

This means there is a problem with PyQt5 installation. Follow this issue for a resolution.

“DLL load failed while importing cv2” on Windows

If you are seeing the following error after installing DepthAI for Windows:

(venv) C:\Users\Context\depthai>python
 Traceback (most recent call last):
   File "C:\Users\Context\depthai\", line 7, in <module>
     import cv2
   File "C:\Users\Context\depthai\venv\lib\site-packages\cv2\", line 5, in <module>
     from .cv2 import *
 ImportError: DLL load failed while importing cv2: The specified module could not be found.

Then installing the Windows Media Feature Pack (here) is often the resolution, as Media Feature Pack must be installed for Windows 10 N editions.

(And more background from OpenCV directly is here)

python3 returns Illegal instruction

This so far has always meant there is a problem with the OpenCV install on the host (and not actually with the depthai library). To check this, run:

python3 -c "import cv2; import numpy as np; blank_image = np.zeros((500,500,3), np.uint8); cv2.imshow('s', blank_image); cv2.waitKey(0)"

If a window is not displayed, or if you get the Illegal instruction result, this means there is a problem with the OpenCV install. The installation scripts here often will fix the OpenCV issues. But if they do not, running

python3 -m pip install opencv-python --force-reinstall

will often fix the OpenCV problem.

Neural network blob compiled with incompatible openvino version

[NeuralNetwork(2)] [error] Neural network blob compiled with incompatible openvino version. Selected openvino version 2020.3. If you want to select an explicit openvino version use: setOpenVINOVersion while creating pipeline

The reason for this error is that depthai can’t resolve the OpenVINO version from the blob. The solution is simple, the user has to specify the OpenVINO version with which the blob was compiled (as mentioned in the error message):

pipeline = depthai.Pipeline()
# Set the correct version:

“realloc(): invalid pointern Aborted” on RPi

On RPi, after running sudo apt upgrade, you might get the error realloc(): invalid pointer\n Aborted when importing cv2 after depthai library. We have observed the same issue, and have found a solution:

  • Downgrade libc6 by running sudo apt install -y --allow-downgrades libc6=2.28-10+rpi1, OR

  • Re-install DepthAI dependencies by running sudo curl -fL | bash

[error] Attempted to start camera - NOT detected!

If you are facing any of the errors above for either Mono Left/Right or Color camera, first try using the latest depthai version (python3 -mpip install depthai -U). If that doesn’t help, there are 2 probable causes:

  • You are using OAK FFC and a camera sensor that isn’t supported by default, so you should use a different branch, see docs here.

  • A camera got disconnected during the shipping. This has been reported only a handful of times, but it’s possible.

The solution here is to open up the enclosure and re-attach the connector to the camera, see the image here for the OAK-D (left mono camera not detected).

[error] input tensor exceeds available data range

[NeuralNetwork(3)] [error] Input tensor '0' (0) exceeds available data range. Data size (6336B), tensor offset (0), size (6912B) - skipping inference

This error is usually thrown when we use NNData message and we don’t provide the amount of bytes that the NN model expects for the inference. For example, in the error above, the NN model expects 6912 bytes (48x48x3), but only 6336 bytes were sent to it.

Converting YUV420 to CV2 frame

If you try to convert YUV420 frame to CV using ImgFrame’s .getCvFrame() method, you might come accross the error below:

cv2.error: OpenCV(4.6.0) d:\a\opencv-python\opencv-python\opencv\modules\imgproc\src\color.simd_helpers.hpp:108: error:
(-215:Assertion failed) sz.width % 2 == 0 && sz.height % 3 == 0 in function
'cv::impl::`anonymous-namespace'::CvtHelper<struct cv::impl::`anonymous namespace'::Set<1,-1,-1>,
struct cv::impl::A0xe823dd8f::Set<3,4,-1>,struct cv::impl::A0xe823dd8f::Set<0,-1,-1>,1>::CvtHelper'

The culprit of the error is that OpenCV requires YUV420 width to be divisible by 2, and height to be divisible by 3. A simple example that will crash:

import depthai as dai
import cv2

pipeline = dai.Pipeline()
cam = pipeline.createColorCamera()

manip = pipeline.createImageManip()
# manip.initialConfig.setFrameType(dai.RawImgFrame.Type.BGR888p)
manip.initialConfig.setResize(600, 451)

xout = pipeline.createXLinkOut()

with dai.Device(pipeline) as device:
    f = device.getOutputQueue('out').get().getCvFrame()
    cv2.imshow("frame", f)

Since isp output is YUV420, it will crash when calling .getCvFrame(). You could either resize the frame to 600x450 (so height is divisible by 3) on line 9, or uncomment the line 8, so frame gets converted to BGR on the device itself.

Got questions?

Head over to Discussion Forum for technical support or any other questions you might have.