Nonlinear Spike-And-Slab Sparse Coding for Interpretable Image Encoding
Fig 8
Results of nonlinear sparse coding using a binary prior on image reconstruction.
The nonlinear sparse coding model if applied to artificial strokes using a fixed narrow slab (SSMCAfix. The two figure columns show image reconstruction results for SSMCAfix with σpr = 0.25 for two different ground-truth stroke numbers (k = 3 and k = 5). SSMCAfix was first trained with fixed σpr and then applied to the data. Reconstructions were computed as described for Fig 7.