Fig. 4From: A learned sparseness and IGMRF-based regularization framework for dense disparity estimation using unsupervised feature learningLearned weights W between the input and the hidden layer in the trained sparse autoencoder. Here, each square block is of size 8×8 which shows the weights between a hidden unit and each input unit. Note that there are 256 hidden and 64 input unitsBack to article page