Empowering Mexican students and educators on TinyML: Running models locally on the XIAO Sense with camera
I had the opportunity as a Seeed Ranger to host two hands on workshops at the offices of Unit Electronics, one workshop in the morning, the second one in the afternoon. Unit Electronics is a well known distributor of electronics and microcontrollers in Mexico.
The workshops where about TinyML vision models running on the XIAO ESP32S3 Sense with camera module.
Exploring TinyML beyond no code
The workshops were centered on how you can use Sense Craft AI platform to collect data, train and deploy a vision model to the XIAO ESP32S3 Sense. The platform makes it very simple to get started with TinyML model deployment, but I also wanted to teach the participants a bit deeper, showing and explaining what is exactly happening on each step:
- Data collection: Participants learned how to gather a good dataset for jobs like image classification or object detection. We talked about the quality, diversity and labeling tools involved to ensure a successful project.
- Model Training in SenseCraft AI: Once the images where captured (each participant decided what they wanted to detect) we trained models straight to the XIAO using SenseCraft AI, showing people that to make your first project you can rely on good tools that make all this process easier. I took the time to explain about how convolutional neural networks (CNN) work and what needs to be done in order to optimize this models to work inside a microcontroller which has memory, processing and energy limitations.
- Deployment: After training we deployed the models to the device and made sure everyone’s model was working well, if not then we checked their dataset in order to improve it and thus the model.
This approach was very well received, since some participants really appreciated going in deeper on what is happening in order to have more possibilities over what models you can deploy.
Who attended?
In collaboration with Unit Electronics we invited people following Unit, this resulted in an attendance mainly of electrical engineering students and a couple of teachers that want to better understand these technologies to later integrate them into their courses. The perfect audience for this type of projects!
Giveaway
Unit Electronic gave some Seeed Studio kits to some of the participants that were able to win a small dynamic at the end. Really like the idea of people testing something in the workshops and go back home to start working on another project on their own hardware!
Brainstorming Tiny ML applications
Between some of the steps, we were able to stop for a couple of minutes and have a couple of brainstorming sessions on how we could integrate Tiny ML projects into our every day life with something’s like the XIAO Sense. People really got the idea really quickly and knew what they would do for example:
- Smart gate that only allows the correct dog to access a bowl of food with medicine. Some projects have been done with RFID or similar technologies, but a participant is working on her thesis with that project but using only vision models.
- Smart monitoring system that can help you automate your house or office with home assistant. Everything running locally and only data coming out of the device once a certain criterion is met.
Showcasing the SenseCAP Watcher was a great catalyzer to spark more ideas, seeing how different systems, local and cloud processing integrate into one same product, allowed the participants to realize they can deploy hybrid systems leveraging the best parts of each approach.
Reflections
The feedback I received was overwhelmingly positive, many participants valued the fact that we went a bit technical to better understand what’s happening. It was really gratifying hearing people talking about all the ideas they wanted to implement on their microcontrollers, there is a lot of potential on the TinyML sector and seeing people so eager to work on implementing AI algorithms intro their computers was a really cool experience.
About Author
