Tiny-dnn is a lightweight dependency free library. It is extremely easy to get your first boilerplate code running as it requires nearly no additional setup. Since it is specifically suited for Mobile and iOT devices, or broadly speaking, devices with low compute capability, it is implied that the focus of library is more on inference rather than training huge models.
That said, there is a need of shipping some standard models, while paying specific attention to ease of loading pre trained weights in them. This project aims to build a Model Zoo for tiny-dnn, providing pretty standard CNN architectures such as AlexNet, VGG, Inception etc. Moreover, the goal is to make transfer learning easily understandable and implementable with tiny-dnn, as one might need to fine tune a huge model on a low end hardware device.

Organization

Student

Karan Desai

Mentors

  • Taiga Nomi
  • Stefano Fabri
  • Edgar Riba
close

2017