Were you able to find any solution for this. Please let me know. Even i am stuck here. I am gettting invoke failed erroe when i am uploading custom models otherwise it is working fine on sense craft platform.
It sounds bad.
Unfortunately I have not had time yet to test the proposal above.
(I think the official custom model uploading will work on The Grove Vision AI Module V2 after a while also. It would be good to know when. In principle it works for V1.)
Thanks for your promt response, Lets hope it works for the Grove Vision AI Module V2 soon. So custom model thing is working for v1 as of now? Could you please tell me if you tried with Grove Vision AI Module V1 module?
Thank you MatBax. This method can be used to create custom model for V2.
I have not had time to try out method of Seeed_Seraphina yet.
(It was faster to try MaxBat’s method because I already had Edge Impulse model.)
Hello,
I have followed the given steps for my keyword detection model. But I am getting an invoke failed error. Does it only work for computer vision applications?
I had a working model from Edge Impulse, which I downloaded from the dashboard (INT8 quantized format). I then renamed it, compiled it using Vela, and deployed it to the Grove Vision AI V2 board. However, I’m encountering an error: “Invoke failed. Please check device connection.”
and I followed the alphabetical order for the model classes as required.
Also,
I attempted to create a classification model using the instructions provided in your documentation (Deploying models using datasets to groove vision AI v2). I followed every step closely, except that I used the dataset directly instead of using Roboflow.
The model trained successfully without any errors, but after deployment, it does not function as expected. For context, it is an MNIST digit classification model. However, the output I’m receiving after deployment is incorrect for example, I get constant result 8:99, which doesn’t make sense as I did not show it any digit to recognize.
So I trained a class model (30 pics) for ON state, and 30 for OFF state and it didn’t work correctly until I added the the Unknown class (30 pics) random stuff. Once rebuilt it worked as expected. Do you have class with random digits or blank digits, maybe try that.
The example was the dev board on or off…
Hey @Pj_Glasso,
Thanks for your reply! I added the “unknown” class to my existing classes, and it’s working now. I really appreciate your help.
However, I tried applying the same approach to keyword detection, but it doesn’t seem to be working. Do you have any idea what might be going wrong in this case?
Awesome, SO I haven’t tried the keyword model yet but will in the near future. I saw a seeed video I think on it and it was pretty accurate but I “do” believe there IS some tricks to getting it to go good , just like the class demo…Sadly I don’t know what they are YET!
PJ 
SO I haven’t tried the keyword model yet but will in the near future. I saw a seeed video I think on it and it was pretty accurate but I “do” believe there IS some tricks to getting it to go good , just like the class demo…Sadly I don’t know what they are YET! 