This repository is deals with road and non-road classification, I did it in Ubuntu14.04, and I write some command to .bat files, so it was the same function in windows. In this repository, it contain lmdb file(it was created from my original data), and my CNN's prototxt file, it's also contain my caffemodel, but it was only train 1000 times. The import things is draw.bat, it can draw loss(training/test) and accuracy(test) real time, so it can help you adjust your CNN's params. It was implement in cpp and py. TrainCaffeNet.py is use ImageDate layer to train CaffeNet to classify road and non-road.