Mojo function

conv_transpose_naive

conv_transpose_naive[dtype: DType](output: TileTensor[dtype, address_space=output.address_space, linear_idx_type=output.linear_idx_type, element_size=output.element_size], input: TileTensor[dtype, address_space=input.address_space, linear_idx_type=input.linear_idx_type, element_size=input.element_size], filter: TileTensor[dtype, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type, element_size=filter.element_size], stride: IndexList[3], dilation: IndexList[3], pad_d: IndexList[2], pad_h: IndexList[2], pad_w: IndexList[2])

Implements the ConvTranspose operator from the MO spec.

Parameters:

dtype (DType): Type of the input, output, and kernel tensors.

Args:

output (TileTensor[dtype, address_space=output.address_space, linear_idx_type=output.linear_idx_type, element_size=output.element_size]): Output data tensor that contains the result of the convolution.
input (TileTensor[dtype, address_space=input.address_space, linear_idx_type=input.linear_idx_type, element_size=input.element_size]): Input data tensor from previous layer, with size of (N x H x W x C), where N is the batch size, C is the number of channels, and H and W are the height and width.
filter (TileTensor[dtype, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type, element_size=filter.element_size]): The weight (kernel) tensor, with size of (kH x kW x M/groups x C), where C is the number of channels, kH and kW are the height and width of the kernel, and M is the number of feature maps.
stride (IndexList[3]): Stride along each spatial axis.
dilation (IndexList[3]): Dilation value along each spatial axis of the filter.
pad_d (IndexList[2]): Padding in depth dimension.
pad_h (IndexList[2]): Padding in height dimension.
pad_w (IndexList[2]): Padding in width dimension.