The implementation of our TMM paper "Deep Learning for Surface Material Classification Using Haptic and Visual Information" prerequisite 0. the code is tested under ubuntu14.04 1. Caffe and matcaffe should be compiled at external/caffe folder. 2. the pretrained Caffenet should be downloaded and placed at models/VisualNet_10fold. 3. the TUM material surface dataset (i.e. the "Accel/" folder and "Image_Database/" folder in file "LMT_TextureDB_1.2.zip") should be unpacked at dataset/TUM. Usage: run startup.m to set path enveriment. run caching_haptic.m and caching_image.m to prepare data run HapticNet_2wide_train_10fold.m to train HapticNet run VisualNet_train_10fold.m to train VisualNet run FusionNet_train_10fold.m to train FusionNet-FC2 run FusionNet_69dim_train_10fold to train FusionNet-FC3 run Compare_TCNN_10fold.m to train VisualNet-TCNN run FusionNet_TCNN_train_10fold.m to train FusionNet-FC2-TCNN run FusionNet_TCNN_train_69_10fold.m to train FusionNet-FC3-TCNN notice all the training is performed with ten-fold cross validation, so they might be extremely slow. One can modify the code to perform a one-fold training.