pyiqa.archs.unique_arch

LIQE Model

github repo link: https://github.com/zwx8981/UNIQUE

Cite as: @article{zhang2021uncertainty,

title = {Uncertainty-aware blind image quality assessment in the laboratory and wild}, author = {Zhang, Weixia and Ma, Kede and Zhai, Guangtao and Yang, Xiaokang}, journal = {IEEE Transactions on Image Processing}, volume = {30}, pages = {3474–3486}, month = {Mar.}, year = {2021}

}

Module Contents

pyiqa.archs.unique_arch.default_model_urls

class pyiqa.archs.unique_arch.Normalize(mean, std)

Bases: torch.nn.Module

Channel-wise normalization module.

forward(x)

class pyiqa.archs.unique_arch.BCNN(thresh=1e-08, is_vec=True, input_dim=512)

Bases: torch.nn.Module

Bilinear CNN pooling block used in UNIQUE.

forward(x)

class pyiqa.archs.unique_arch.UNIQUE

Bases: torch.nn.Module

UNIQUE no-reference image quality model.

Parameters:

• pretrained (No runtime arguments. The model loads the default)

• checkpoint. ('mix')

forward(x)

Predict quality score using UNIQUE.

Parameters:

x (torch.Tensor) – Input tensor with shape (N, 3, H, W).

Returns:

Predicted mean quality score with shape (N,).

Return type:

torch.Tensor