I have trained a multi-class image classification model using MobileNetV2 on Edge Impulse and even after successfully building the .uf2 file, its not working when I am trying to run edge-impulse-run-impulse or even minicom.
Hi @chinya07
Can you give some detail on the configuration you made prior to deployment? and also the project ID? It is possible that the configuration lead to an out of memory issue. Thanks!
Best
Eoin
Hi @Eoin, I have around 2.5k images with each image belongs to 1 of 2 classes. The images are 160X160X3 and I am trying to train them with MNV2 pretrained base model. I didn’t use the EON Tuner for optimizations. I am not getting any memory overflow error for int8 model and its getting built to final .uf2 file.
The project id is 123665.
Hi @chinya07
It looks like your model is too big for the current firmware. Please use our EON Compiler feature that will decrease your resources requirement, and use an optimized model (int8)
I tested your model, and with the settings above, it runs on the Vision Ai module.
Best regards,
Mateusz