Relationship between HOF, MBH, convolution layer and sparse coding

HOF  VS ConvNet: HOF descriptor  based on  the histograms of optical flow orientations. But convolution network also can do this: given a displacement field input, using single convolution layer that containing orientation-sensitive filters, followed by the rectification and pooling layer.

MBH VS ConvNet: MBH can be computed by the histograms of flow gradient orientations; this corresponds to adding an additional convolutional layer, making it a two-layer representation. But ConvNet repeat this by statcking two conv layers.

Sparse coding VS ConvNet: ConvNet can combine a single convolutional and rectification layer to form a sparse coding.