Hello all! I've only recently received the OAK-1 for a research project. Like a kid at Christmas, I've been super excited to try it out with a custom-trained model.

I followed the "Easy_TinyYOLOv3_Object_Detector_Training_on_Custom_Data.ipynb", great tutorial! However when testing it with my OAK-1 I get this error:

[14442C1051D14FD000] [301.575] [DetectionNetwork(2)] [warning] Network compiled for 5 shaves, maximum available 13, compiling for 6 shaves likely will yield in better performance
[14442C1051D14FD000] [301.887] [system] [critical] Fatal error. Please report to developers. Log: 'XLinkOut' '221'
Traceback (most recent call last):
File "C:\Users\adam\Documents\DepthAI Demo\depthai-main\depthai_demo.py", line 637, in <module>
pv.prepare_frames()
File "C:\Users\adam\Documents\DepthAI Demo\depthai-main\depthai_demo.py", line 87, in prepare_frames
frame = queue.tryGet()
RuntimeError: Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: 'color' (X_LINK_ERROR)'

  • erik replied to this.

    Hello Kialo, we apologize for this first impression of our notebooks. Just recently (3 days ago) I went through this tutorial and it worked as expected for me (discussion here, where you can also find script/blob). Could you share your blob model? I can give it a try with the script I mentioned in the other discussion.
    And at Luxonis we encourage our users to try Roboflow for training (where users can also get much better ML/AI support than they could here) and they have a course for the depthai platform.
    Thanks, Erik

        Hello erik, thank you for the information. I'll take a look at that discussion.

        Here is my blob model:

            erik I've requested access to train the model with Roboflow.

            No need to apologize for the notebooks, they have helped me tremendously with getting a fundamental understanding of all the requirements for depthAI and my OAK-1 device.

              Hello Kialo, I tried your model and can confirm that it's not working as expected. It might be better to try out the yolov4, I don't think we have had any problems with those blobs. Thanks, Erik

                  erik Thanks for trying it, it was only a test to prototype the shapes anyway. I've now set up a more real-world testing environment. Then captured the photos directly on the OAK-1, so that it may be as close to the intended application. I'll now be using Roboflow to train the blob. I'll report back once that's completed.