With the ever-increasingly performant mobile models that allow AI models to be embedded directly within devices itself, applications that have AI capabilities on-device are increasing appealing to the development market.
There are many use cases for such models, particularly using models that are capable of the following:
- Computer Vision (e.g. object detection, instance segmentation, etc.)
- Natural Language Processing (e.g. LLMs)
- Video Processing (e.g. video to language frameworks)
- Machine Translation
And many more.
Given the wide capability of models nowadays, the following are mainly tested for the respective capabilities.
- YOLOv8 to YOLOv12 TFLite models
- SmolVLM
- PaliGemma
- Gemma
The applications for testing are mainly developed using Kotlin with Jetpack Compose.
There are future plans of porting over to Kotlin Multiplatform, but for the sake of compatibility with native Android, Kotlin is preferred.
In terms of libraries used, the following are the main ones:
- TensorFlow Lite
- Google MLKit