This was a small mobile application used as a component of FastCart in a hackathon at The Center of Financial Technology in Tomsk, Russia.
The vast majority of this project was written by avihay, please check out his blog post and repo for a really cool traffic light recognition example.
Clone the project (recursivly) to get Caffe as well:
$ git clone --recursive https://github.com/samsheff/FastCartMobile
Fetch OpenCV2 by running the script:
./download_dependencies.sh
Download our model file from here and place it in the model folder in XCode
Open the XCode project and run it
Using your own model is simple. Simply replace the files under the model directory with the ones your downloaded from DIGITS
-
Caffe can be installed on various operating systems with or without a virtual environment. I do however recommend you to install it on a Ubuntu 14.04 machine without any virtualization (e.g. Docker/VirtualBox). It is quite complex to make it happen and you will be wasting time solving compilation problems instead of having fun.
-
Under normal project setup, the optimization flag for your DEBUG build setting is set to None. This will cause the prediction code to run significantly slower than your production build. If you are testing your applications performance, make sure you change the optimization flag to Fastest [-Os]. On my iPhone 7 device I experienced a X3 performance increase.
Thanks to avihay, Aleph7 and noradaiko for the preliminary work that powered this example.