mmagic.models.editors.ttsr.lte

Module Contents

Classes

LTE	Learnable Texture Extractor.

class mmagic.models.editors.ttsr.lte.LTE(requires_grad=True, pixel_range=1.0, load_pretrained_vgg=True, init_cfg=None)

Bases: mmengine.model.BaseModule

Learnable Texture Extractor.

Based on pretrained VGG19. Generate features in 3 levels.

Parameters

• requires_grad (bool) – Require grad or not. Default: True.

• pixel_range (float) – Pixel range of feature. Default: 1.

• load_pretrained_vgg (bool) – Load pretrained VGG from torchvision. Default: True. Train: must load pretrained VGG. Eval: needn’t load pretrained VGG, because we will load pretrained LTE.

• init_cfg (dict, optional) – Initialization config dict.

forward(x)

Forward function.

Parameters

x (Tensor) – Input tensor with shape (n, 3, h, w).

Returns

Forward results in 3 levels.

x_level3: Forward results in level 3 (n, 256, h/4, w/4). x_level2: Forward results in level 2 (n, 128, h/2, w/2). x_level1: Forward results in level 1 (n, 64, h, w).

Return type

Tuple[Tensor]