Following Visual Anomaly for OpenMV RT1062

#1

I made a Visual Anomaly project Visual Anomaly - Dashboard - Edge Impulse with a deploy to TDA4VM and everything worked perfectly.

Now I want to deploy an OpenMV Cam RT1062 Library, but:

    1. The library zip includes an Image Classification Python script ei_image_classification.py
    1. Label file has one label: “no anomaly”
    1. The TFLite model returns outputs between 0 and 1

{“x”:160, “y”:60, “w”:20, “h”:20, “output”:0.8000001}
{“x”:180, “y”:100, “w”:40, “h”:40, “output”:0.8980392}
{“x”:120, “y”:180, “w”:40, “h”:40, “output”:0.9411765}
{“x”:0, “y”:80, “w”:20, “h”:20, “output”:0.8901961}

while the deploy to TDA4VM returns, as expected, the anomaly scores matrix:

[{“height”:19,“label”:“anomaly”,“value”:4.347522258758545,“width”:19,“x”:0,“y”:0},{“height”:19,“label”:“anomaly”,“value”:3.812622547149658,“width”:19,“x”:19,“y”:0},{“height”:19,“label”:“anomaly”,“value”:3.6791884899139404,“width”:19,“x”:38,“y”:0},{“height”:19,“label”:“anomaly”,“value”:3.572544574737549,“width”:19,“x”:57,“y”:0}, etc

Is there some way to deploy Visual Anomaly for OpenMV?

Thanks again

CreateImpulse
CreateImpulse1920×1098 89.9 KB

1 Like
#2

Hi @Bandini

We are working on a fix ATM for the OpenMV firmware, it should be live by tomorrow. Can you try again tomorrow?

Best

Eoin

#3

Sure. Thanks for the answer @Eoin

#4

@Bandini Were you able to get this to work? Also, I don’t know how to print out the anomaly score, have tried both with deploying as library and as firmware, but might be I’m missing something obvious.

#5

Hi @ThomasVikstrom
The OpenMV integration runs plain Tflite files. This means everything handled outside of Tflite (like the AD part of FOMO-AD) is not supported

1 Like
#6

Aha, ok, back to the drawing board :slight_smile: Thx!

#7

I believe AD is still under revision for support @Bandini.

Check with David on the expert call about ETA for the OpenMV but the TDA4VM wont be supported any time soon.

Best

Eoin

1 Like
#8

I will check that. Thanks Eoin.