Hi everyone ! I'm working on a QR detection project, I've followed the depthai-experiments/gen2-qr-code-scanner tutorial and it works perfect... my question is: is there a BlobPath (NN) for Datamatrix or BARCode detection?

If not, how could I train the neural network for Datamatrix or BARCode, thanks.

  • erik replied to this.

    Hi FabianRamirez ,
    We don't have any example with barcode NN model, nor do we have such model in depthai model zoo. I'd imagine it would be best to search for weights online instead of training your own barcode model. From quick search, Google's ML Kit should have barcode detection model inside, looking online it would be possible to get the weights from ML Kit and then convert/deploy them to the device. THoughts?
    Thanks, Erik

    Hi Erik, thanks for response…I don't know how to do this, I think I should continue investigating, I'm going to look at this information you refer to on Google models... but I don't understand adjusting the weights, looking at the ML Kit website, it's only for Android and iOS.. =(

    • erik replied to this.

      Hi FabianRamirez , I looked into it, if you build the android app with mlkit enabled and unzip the app, you can find the tflite model weights in there (see below). For conversion and decoding might not be that trivial though. I was checking for mlkit_barcode_models, it's mobilenet, but it doesn't have NMS, so you'd need to implement it on the host (likely). I might look into this in the upcoming days. THoughts?

      Hi Erik, I will continue investigating, it is interesting and appropriate for my project to be able to read barcodes or datamatrix, the camera OAK lends itself to what I need, thanks for the help, if you have any other suggestion I would appreciate it, regards!

      • erik replied to this.

        Hi FabianRamirez ,
        Sounds good, we will update if there are any updates, perhaps we will try to compile the model and deploy it to the OAK ourselves as well. Please also let us know if you make any progress on this🙂
        Thanks, Erik