Tagging stable version 2018.12.20

Tagging stable version 2018.12.20

important bug fix on new mobiles

• fixed "access deined" error when detecting platform info on new mobiles with Android 7+ (Samsung S8+, Huawei Mate, etc)

New version with support for OpenCL DNN engines

This new version supports OpenCL when crowdsourcing benchmarking and optimization of various DNN algorithms.

Tagging stable V2.6 version

Main various bug fixes - now stable!

text updates

Minor update to V2.0 with text update.

Major redesign

• Daniil Efremov considerably redesigned this version to be much more user-friendly and intuitive.
• Grigori Fursin provided preparation of crowd-benchmarking and crowd-tuning scenarios for this application (such as compilation of various DNN libraries for Android devices) completely automatically via Collective Knowledge: https://github.com/ctuning/ck-crowd-scenarios .
• Public results are available at http://cKnowledge.org/repo (select crowd-benchmark DNN libraries using mobile devices).
• We would like to thank Anton Lokhmotov, Dmitry Savenko and the community for testing, feedback and future directions!

v1.3

This beta version powered by Collective Knowledge can:

• Support basic DNN benchmarking (init+recognition) for Android 5+ (time and accuracy)
• Support for DeepScale Squeezenet 1.0 and 1.1, BVLC CaffeNet and AlexNet, and BVLC GoogleNet
• Share results via http://cknowledge.org/repo (see "crowd-benchmark Caffe library using mobile devices" scenario)
• Share mispredictions (images and correct labels)
  NOTE: Don't forget to press 'Update' to get latest crowd-scenarios!