mmagic.models.editors.inst_colorization.weight_layer

Module Contents

Classes

WeightLayer

Weight layer of the fusion_net. A small neural network with three

Functions

get_norm_layer([norm_type])

Gets the normalization layer.

mmagic.models.editors.inst_colorization.weight_layer.get_norm_layer(norm_type='instance')

Gets the normalization layer.

Parameters

norm_type (str) – Type of the normalization layer.

Returns

normalization layer. Default: instance

Return type

norm_layer (BatchNorm2d or InstanceNorm2d or None)

class mmagic.models.editors.inst_colorization.weight_layer.WeightLayer(input_ch, inner_ch=16)

Bases: mmengine.model.BaseModule

Weight layer of the fusion_net. A small neural network with three convolutional layers to predict full-image weight map and perinstance weight map.

Parameters

• input_ch (int) – Number of channels in the input image.

• inner_ch (int) – Number of channels produced by the convolution. Default: True

resize_and_pad(feauture_maps, info_array)

Resize the instance feature as well as the weight map to match the size of full-image and do zero padding on both of them.

Parameters

• feauture_maps (tensor) – Feature map

• info_array (tensor) – The bounding box

Returns

Feature maps after resize and padding

Return type

feauture_maps (tensor)

forward(instance_feature, bg_feature, box_info)

Forward function.

Parameters

• instance_feature (tensor) – Instance feature obtained from the colorization_net

• bg_feature (tensor) – full-image feature

• box_info (tensor) – The bounding box corresponding to the instance

Returns

Fused feature

Return type

out (tensor)