Skip to content

RenderScript based implementation of Convolutional Neural Networks for Android phones

Notifications You must be signed in to change notification settings

mtmd/Mobile_ConvNet

Repository files navigation

Android and RenderScript Based Convolutional Neural Networks

The project implements SqueezeNet on Android phones. We use RenderScript efficiently to minimize the execution time. I have tested the project on three different Android phones (Nexus 5, Nexus 6P and Samsung Galaxy). The achieved speedups on these platforms are 310X, 133.89X, and 59.54X respectively. The results are submitted to a conference as a paper which is under review.

Documentation

As the documentation overlaps with the content of the paper, we will publish them when we get the result of submission. However, if you are interested in working on the project, please contact me for more information.

Email

mmotamedi@ucdavis.edu

Citing

If this project helped your research, please kindly cite our work.

Motamedi, Mohammad, Daniel Fong, and Soheil Ghiasi. "Cappuccino: Efficient Inference Software Synthesis for Mobile System-on-Chips." arXiv preprint arXiv:1707.02647 (2017).

About

RenderScript based implementation of Convolutional Neural Networks for Android phones

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published