YoloV5 - Wrong output layer mask

medonni

Hi all!

I'm trying to implement my custom yolov5 model into the oak-d-lite camera but i stumbled upon an issue where i didnt find much help.

The model was trained and i got an .onnx file, which worked pretty well directly through openCV. I did the conversion into .blob using http://blobconverter.luxonis.com/ and obtained my .blob file. Also i've made the corresponding JSON config file as instructed, changing the number of classes and their labes as shown on the bottom of the page here https://github.com/luxonis/depthai-ml-training/blob/master/colab-notebooks/YoloV5_training.ipynb

Now onto the issue.

I've loaded up my blob and cfg into the program and this error keeps popping out.
[1844301041DDC21200] [33.853] [DetectionNetwork(1)] [error] Mask is not defined for output layer with width '6'. Define at pipeline build time using: 'setAnchorMasks' for 'side6'.

Any help is appriciated!

erik

Hi medonni ,
I believe you need to set anchors as noted in the notebook:

Could you try that out?
Thanks, Erik

medonni

erik Hey Erik!

I managed to fix the issue recently. It was an issue on my end where i did not do the necessary pruning as instructed in the notebook. I thought that just any regular .ONNX would work.

After doing the pruning, and generating another .blob it worked fine.

erik

Great that you got it working and thanks for circling back medonni !