The Argus Client is an Android app that leverages remote, large-scale neural networks to provide fast and reliable image recognition on low-powered devices. The client sends the image in a compressed, native YUV format and accepts replies in JSON format. It supports bounding boxes, class labels, and mask pedictions, and allows users to actively provide feedback training to the model.

Testing was done using a server running a CNN with GPU-acceleration enabled. The app is written in Java and C++, using Android API and Google's libyuv. Image processing was done using OpenCV, and the test servers ran custom neural networks as well as Caffe.