Pixe-wise classification to indentify various regions like tree, ground, sky, water etc. in outdoor images

Machine learnig framework used = caffe with other libraries

Total classes = 10

The mapping is as follows:

{'building': 1, 'dirt': 2, 'foliage': 3, 'grass': 4, 'human': 5, 'pole': 6, 'rails': 7, 'road': 8, 'sign': 9, 'sky': 10} 0 means unlabelled and can be ignored. Note: Unfortunately complete data can't be open-sourced as it belongs to AirLab CMU, but I do have included a sample with permission

channels.py Extracting the six channels of RGB image written in python

makehdf5.py File to convert the stored data and label in mat to hdf5 format is

solver.prototxt trainer.prototxt deploy.prototxt caffe files that contains a new network architecture for the current approach, still improving

segment.py implements caffe model

runtrain.sh run training using which creates log also

A short report.pdf a brief idea of the work and things accomplished