tiny_diff.modules.conv

Functions

zero_module(module)

Zero out the parameters of a module and return it.

Classes

ConvBlock([scale_factor, stride, padding, ...])	See https://paperswithcode.com/method/fixup-initialization.
PreNormActConvBlock([scale_factor, stride, ...])	Conv blcok with BN and Nonlinearity before conv instead of after.
UpsampleConvBlock([scale_factor])	Convolutional block that upscales the output.