networks

Networks for AID

WideNet

class indexedconv.nets.aid.WideNet(n_out)

ResNet like Network from HexaConv paper (Z²).

Parameters:n_out (int) – Number of features after last convolution.

WideNetIndexConvIndexPool

class indexedconv.nets.aid.WideNetIndexConvIndexPool(index_matrix, camera_layout, n_out)

ResNet like Network from HexaConv paper implemented with indexed convolutions and pooling.

Parameters:
  • index_matrix (torch.Tensor) – The index matrix corresponding to the input images.
  • camera_layout (str) – The grid shape of the images.
  • n_out (int) – Number of features after last convolution.

WideNetMasked

class indexedconv.nets.aid.WideNetMasked(n_out)

ResNet like Network from HexaConv paper implementing masked convolutions.

Parameters:n_out (int) – Number of features after last convolution.

Networks for CIFAR

WideNet

class indexedconv.nets.cifar.WideNet

ResNet like Network from HexaConv paper (Z²).

WideNetIndexConvIndexPool

class indexedconv.nets.cifar.WideNetIndexConvIndexPool(index_matrix, camera_layout)

ResNet like Network from HexaConv paper implemented with indexed convolutions and pooling.

Parameters:
  • index_matrix (torch.Tensor) – The index matrix corresponding to the input images.
  • camera_layout (str) – The grid shape of the images.

WideNetIndexConvIndexPoolRetina

class indexedconv.nets.cifar.WideNetIndexConvIndexPoolRetina(index_matrix, camera_layout)

ResNet like Network from HexaConv paper implemented with indexed convolutions (retina like kernel) and pooling.

Parameters:
  • index_matrix (torch.Tensor) – The index matrix corresponding to the input images.
  • camera_layout (str) – The grid shape of the images.

Network for MNIST

GLNet2HexaConvForMnist

class indexedconv.nets.mnist.GLNet2HexaConvForMnist(index_matrix)

Network with indexed convolutions and pooling (square kernels). 2 CL (after each conv layer, pooling is executed) 1 FC

Parameters:index_matrix (torch.Tensor) – The index matrix corresponding to the input images.