Hi,
I used to prepare ML model (object detection/classification) using Edge Impuls and Sensecraft. Is Ameba Mini also can be trained to detect or recognize custom objects? If yes how?
Hi @Zul77 ,
For AMB82 Mini, it has a built-in NN engine that can perform tasks like object detection, face detection recognition etc.
Before loading the model to AMB82 Mini, you have to convert your model to a format that AMB82 Mini accepts. We have 2 kinds of conversion tools: online and offline. You can refer to this page: https://www.amebaiot.com/en/ameba-arduino-summary/ under “Custom AI model”.
You can train your own model and deploy onto AMB82 Mini using 2 methods,
- Load using SD card
- Load using flash
Find out how to deploy models here: Customized AI Model Installation Guide — Ameba Arduino AIoT Documentation v1.1 documentation
May I know if you would like to use Arduino SDK or FreeRTOS SDK?
Thank you. I hope there will be more tutorial for Ameba Mini, esspecially how to train custom objects.
Hi @Zul77
For example, you can refer to How to Train a YOLOv7 Model on a Custom Dataset for your custom dataset
Please train the model by tiny architecture and execute Re-parameterization after training to deploy it on AMB82 mini
Thank you, I will try.
Hi, thanks for you reply. I want to use Arduino SDK. I have trained my model using CNN and I have converted the .H5 model using the tool in the website. Now I have the .nb file.Can you guide me how to use the .nb file in Arduino sdk please?
Hi @Zul77, there are 2 ways to load NN model onto AmebaPro2.
- Flash
- SD Card
You can refer to the above guide to load your customised model.
1 Like
Thanks for your reply. I want to load the NN model using the flash method. I have named my .nb file ‘img_classification.nb’. I the link that you shared above, it does not mention how to name the file for the object classification project.
Hi @Zul77 ,
I have replied the above in Custom Image Classification - #4 by pammyleong