Too much metadata error

Daniel_UWA

I have two model to run on the oak camera.

One model works fine, the other model, which is the same size has the following error:

[XLinkOut(3)] [error] Message has too much metadata (125335B) to serialize. Maximum is 51200B. Dropping message

What could be causing this error?

jakaskerl

Hi Daniel_UWA
This could be caused by a number of factors, but it's likely related to the size of the model you're trying to run, or the amount of data it's producing. If your model is producing a large amount of metadata, you may need to adjust your model or the way it's being processed to reduce the amount of metadata being generated.

Does the model run fine on it's own - without the first model?

Thanks,
Jaka

Daniel_UWA

jakaskerl

Just to clarify, I have two models that I am testing to see which one is better. I am trying each one separately on a test video to see which is better. They are both Yolo v5 (and prunned, etc. according to the docs) just trained on different datasets. One model works, the other does not.

What are some options to reduce the amount of metadata?

jakaskerl

Hi @Daniel_UWA
Any chance you can upload the models so our ML team can check? Metadata error is usually a sign of incorrectly converted models.

cc @Matija

Thanks,
Jaka

MatevzMorato

Hi @Daniel_UWA.

Metadata in this case relates to the detections, so a lot of metadata means a lot of detections >100.

Is this expected in your case?

You can try increasing the confidence threshold with setConfidenceThreshold(threshold)

Though if a high number of detections is not what you expect, there is likely a problem with the model.