Custom AI/ML model for Edge Computer Vision

I’m co-author on a paper ‘CupidShuffle’ that modifies MobileNet embedding to to use Transformers, result is an efficient embedded computer vision model targeting edge/IOT.

The core idea and implementation were performed by co-author Matt, I helped with reproducibility and presentation. (source code)

The point was some good fun in showing that The Transformer and self attention isn’t just for GenAI / LLMs / Diffusion, and can work efficiently in computer vision (which was not true at the start!).

also see MobileFormer

tools: 

PyTorch, Apache TVM, C++, Python