calibrate.py problem

[olivergh1708]

Describe the bug
I have two luxonis cameras (OAK-D and OAK-D-LITE) and I am having problems using the calibrate.py script with both of them.

• OAK-D camera:
  The RGB camera is blurry and cannot identify the charuco board.

python3 calibrate.py -s 3.85 --board OAK-D -nx 35 -ny 20
[OK] OpenCV version 4.5.5 is correct.
Cam: color and focus: True
Cam: left and focus: False
Cam: right and focus: False
Saving dataset to: dataset
Sensor name for color is IMX378
Sensor name for left is OV9282
Sensor name for right is OV9282
Starting image capture. Press the [ESC] key to abort.
Will take 39 total images.
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
Start capturing...
new minimum: {'ts': 0.0333479999999895, 'indicies': {'color': 0, 'left': 0, 'right': 0}} min required: 0.2
Time stamp of color is 1 day, 17:19:03.736332
Markers count ... 0
Total markers needed -> 140
Status of color is False
Time stamp of left is 1 day, 17:19:03.736166
Markers count ... 271
Total markers needed -> 140
py: Saved image as: dataset/left/p0_0.png
Status of left is True
Time stamp of right is 1 day, 17:19:03.769513
Markers count ... 299
Total markers needed -> 140
py: Saved image as: dataset/right/p0_0.png
Status of right is True
py: Capture failed, unable to find chessboard! Fix position and press spacebar again

• OAK-D-LITE
  In this case I couldn't even open the video stream.

python3 calibrate.py -s 7.70 --board OAK-D-LITE -nx 17 -ny 9 -v3
[OK] OpenCV version 4.5.5 is correct.
Cam: color and focus: True
Cam: left and focus: False
Cam: right and focus: False
Saving dataset to: dataset
Sensor name for color is IMX214
Sensor name for left is OV7251
Sensor name for right is OV7251
Starting image capture. Press the [ESC] key to abort.
Will take 39 total images.
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
Traceback (most recent call last):
  File "/home/it_poc/workspaces/oliver_aruco/depthai/calibrate.py", line 1255, in <module>
    Main().run()
  File "/home/it_poc/workspaces/oliver_aruco/depthai/calibrate.py", line 1246, in run
    self.capture_images_sync()
  File "/home/it_poc/workspaces/oliver_aruco/depthai/calibrate.py", line 920, in capture_images_sync
    combinedImage = np.hstack((combinedImage, subImage))
  File "/home/it_poc/.local/lib/python3.10/site-packages/numpy/core/shape_base.py", line 359, in hstack
    return _nx.concatenate(arrs, 1, dtype=dtype, casting=casting)
ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 0, the array at index 0 has size 239 and the array at index 1 has size 240

Could you give me some advice on how I could manage to do a calibration of the camera please.

[effgee]

I have the same issue with my oak-D-lite units.
Is anyone form Luxonis active here ?

[themarpe]

CC: @MaticTonin for input on this^

[brmarkus]

Can both of you @effgee and @olivergh1708 provide information about your environment? What OS (looks like Linux, which kernel version?) are you on, which Python version, which device firmware(s) is installed on the cameras?
Has calibration been working before, or is it the first time you try it?
How have you connected the cameras, directly to an USB-C plug, no adapter, no converter, no active/passive USB hub?

Have you tried other positions (camera <-> display), other distances (camera <-> display)?

Just for the records: https://docs.luxonis.com/hardware/platform/depth/manual-calibration

[MaticTonin]

Hi.
For the RGB camera, you must manually adjust the focus with adding the -rlp rgb=ideal_lens_position_. To get that value, it depends on your distance from camera to target, what would be ideal.

For issues with LITE, seems like there is a mismatch between image size and padding, can you try this branch?
#1246

[Seneral]

Hi. For the RGB camera, you must manually adjust the focus with adding the -rlp rgb=ideal_lens_position_. To get that value, it depends on your distance from camera to target, what would be ideal.

For issues with LITE, seems like there is a mismatch between image size and padding, can you try this branch? #1246

I'm running into the same issue (in trying to upgrade my EEPROM from version 6 to 7 - ideally I'd like to keep the factory calibration).
I tried your branch (applied patch onto current master), and it did fix the issue when capturing images, but after processing, it encountered a very similar issue again (plus device comms issues):

...
Use DEFAULT model
Filtering 0.10019469261169434s
Filtering takes: 0.3127291202545166
Use DEFAULT model
Each filtering 0.10159683227539062
Each calibration 0.20341777801513672
Each filtering 0.0994257926940918
Each calibration 0.24358105659484863
Each filtering 0.09871625900268555
Each calibration 0.08195042610168457
Whole procedure: 0.8293280601501465
Reprojection error of left: 0.331447823171523
[2026-06-12 14:13:55.661] [depthai] [error] Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: '__x_8__output' (X_LINK_ERROR)'
[2026-06-12 14:13:55.661] [depthai] [error] Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: '__x_0_out' (X_LINK_ERROR)'
[2026-06-12 14:13:56.870] [ThreadedNode] [error] Node threw exception, stopping the node. Exception message: Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: '__x_0_out' (X_LINK_ERROR)'
[2026-06-12 14:13:56.870] [ThreadedNode] [error] Node threw exception, stopping the node. Exception message: Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: '__x_8__output' (X_LINK_ERROR)'
Device closed in exception..
all the input array dimensions except for the concatenation axis must match exactly, but along dimension 0, the array at index 0 has size 479 and the array at index 1 has size 480
Traceback (most recent call last):
  File "/mnt/temp/BUILD/depthai/./calibrate.py", line 1063, in calibrate
    status, result_config = stereo_calib.calibrate(
                            ~~~~~~~~~~~~~~~~~~~~~~^
                                self.board_config,
                                ^^^^^^^^^^^^^^^^^^
    ...<5 lines>...
                                self.args.cameraMode,
                                ^^^^^^^^^^^^^^^^^^^^^
                                self.args.rectifiedDisp) # Turn off enable disp rectify
                                ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/temp/BUILD/depthai/depthai_calibration/calibration_utils.py", line 361, in calibrate
    combinedCoverageImage = np.hstack((combinedCoverageImage, subImage))
  File "/home/louis/.venv/depthai-update/lib/python3.14/site-packages/numpy/core/shape_base.py", line 359, in hstack
    return _nx.concatenate(arrs, 1, dtype=dtype, casting=casting)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 0, the array at index 0 has size 479 and the array at index 1 has size 480

[Seneral]

One further patch (just applying your changes to that problem) and it does finish processing.

I now have the old EEPROM json (version 6) in the tmp with just the two flags modified - thanks for adding that backup with this patch!

I also have the new EEPROM json (though still version 6) written to the resources folder, with modified calibration for just the cameras.
That "new" EEPROM is actually exactly the same as the tmp one: Old calibration, Version 6, modified flags. None of the new calibration.

However, when reading the EEPROM again from the device (with the calibration_dump script), I confirmed it was not even written at all.
Notably, I have no factory calibration on there either, just a user calibration (which IS the factory calibration, I don't think I ever recalibrated).

But now that I have the old EEPROM - is there any way I can manually upgrade the version 6 to version 7 and just rewrite it with the same factory calibration? I just do this because the on-device undistortion stopped working and it seems to be because of this version bump.

[Seneral]

Btw here is the combined patch of your branch and that additional fix, for the current main branch: patch.txt

[Seneral]

What a mess.
So turns out I never backed up the original calibration, the tmp one was already the new calibration. So I only have the new user calibrations since my device apparently never had a factory calibration (likely because it's a Kickstarter model).
So all factory calibration is gone (I do have a backup of distortion coefficients of the RGB camera at least).
I wish I'd known about calibration_dump before trying all this.

Anyway, what's wrong with calibrate.py?
Apart from the padding fixes, it also lacks any upgrade path to move from EEPROM version 6 to 7. The patch above fixes the two boolean flags and changes 6 to 7 in two different ways.
The EEPROM version bump did not work, and was never applied.
The flag updates were reset by the camera itself, when read back they were never applied. Likely because it still read EEPROM version 6 and corrected it, but that's just an assumption.

One (potentially dangerous) way to fix this is to run with "--noInitCalibration " - this will stop using the poisoned EEPROM version 6 and start with a blank EEPROM version 7. Another flag of note is "--debugProcessingMode", though to actually get a JSON file written you need another patch since that's erroneously not written even with above patch.

The other - proper - way to fix this is to actually upgrade the EEPROM properly. I do not know exactly how to do this, only Luxonis knows, and since I don't have a proper V6 and V7 to compare, only borked V7s, I can't help there.
But I can fix the script to at least update the flags and the version number properly. AND WRITE A BACKUP.

TLDR

Here's my patch: final-patch.txt - apply it to the main branch with git apply final-patch.txt.
You can do your calibration normally and it will write a backup, and HOPEFULLY properly upgrade to v7. If it does not, try --noInitCalibration, and it will start with a blank slate v7. With a backup, you can always return to your original v6 with calibration_flash.
If you just want to make sure you have a backup first, use calibration_dump, or run this updated calibrate.py with --debugProcessingMode, which will create the backup and write the new calibration JUST to file, not to the device EEPROM.

[effgee]

To clarify, when I said "same problem here" I was referring to being unable to get video from my oak-d-lite(s). Windows 11, direct USB C cable

The calibration sounds like a nightmare, kudos on your persistence.

[Seneral]

Right, got that, and that problem IS solved.
I just had another problem that might be due to my OAK-D Lite being a kickstarter model (though I'm not sure). It affects all cameras that shipped with EEPROM version 6 instead of 7. There's no autoupdate, and since a few months ago I believe, the enableUndistortion flag (e.g. on-device distortion correction) silently stopped working, I presume because they just stopped supporting the version 6. I doubt they'd do that unless it only affects the very first cameras they produced. I still believe they should have handled that better.
You can just do a quick print("version:", device.readCalibration().eepromToJson().get("version")) to check if your device is affected.
Either way, if you're using an OAK-D lite, the patch I sent above fixes calibration for you.