mmedit.models.base_archs.conv2d_gradfix

Custom replacement for torch.nn.functional.conv2d that supports arbitrarily high order gradients with zero performance penalty.

Module Contents

Functions

no_weight_gradients([disable])
conv2d(input, weight[, bias, stride, padding, ...])
conv_transpose2d(input, weight[, bias, stride, ...])
_should_use_custom_op(input)
_tuple_of_ints(xs, ndim)
_conv2d_gradfix(transpose, weight_shape, stride, ...)

Attributes

enabled
weight_gradients_disabled
_conv2d_gradfix_cache
_null_tensor

mmedit.models.base_archs.conv2d_gradfix.enabled = True

mmedit.models.base_archs.conv2d_gradfix.weight_gradients_disabled = False

mmedit.models.base_archs.conv2d_gradfix.no_weight_gradients(disable=True)

mmedit.models.base_archs.conv2d_gradfix.conv2d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1)

mmedit.models.base_archs.conv2d_gradfix.conv_transpose2d(input, weight, bias=None, stride=1, padding=0, output_padding=0, groups=1, dilation=1)

mmedit.models.base_archs.conv2d_gradfix._should_use_custom_op(input)

mmedit.models.base_archs.conv2d_gradfix._tuple_of_ints(xs, ndim)

mmedit.models.base_archs.conv2d_gradfix._conv2d_gradfix_cache

mmedit.models.base_archs.conv2d_gradfix._null_tensor

mmedit.models.base_archs.conv2d_gradfix._conv2d_gradfix(transpose, weight_shape, stride, padding, output_padding, dilation, groups)