cutcutcodec.core.nn.model.compression.img_cgavaenn.Decoder

class cutcutcodec.core.nn.model.compression.img_cgavaenn.Decoder[source]

Unfold the projected encoded images into the color space.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(lat: Tensor, *, mse: bool = True, gen: bool = True) → Tensor[source]

Apply the function on the latent images.

Parameters

lattorch.Tensor: The projected image in the latent space of shape (n, 256, hl, wl).
mseboolean, default=True: If True, return the mse head result at first position, return None otherwise.
genboolean, default=True: If True, return the generative head result at second position, return None otherwise.

Returns

img_msetorch.Tensor or None: A close image in colorspace to the input image. It is as mutch bijective as possible than VariationalEncoder. New shape is (n, 256, 160+hl*32, 160+wl*32) with value in [0, 1].
img_gentorch.Tensor or None: A beautifull image in colorspace, don’t match very accurately to the original. It can be extrapolated in order to reinvent details. New shape is (n, 256, 160+hl*32, 160+wl*32) with value in [0, 1].

Examples

>>> import torch
>>> from cutcutcodec.core.nn.model.compression.img_cgavaenn import Decoder
>>> decoder = Decoder()
>>> mse, gen = decoder(torch.rand((10, 256, 1, 3)))
>>> mse.shape
torch.Size([10, 3, 192, 256])
>>> gen.shape
torch.Size([10, 3, 192, 256])
>>>