pyiqa.archs.deepdc_arch

DeepDC: Deep Distance Correlation as a Perceptual Image Quality Evaluator

Reference: @article{zhu2024adaptive,

title={DeepDC: Deep Distance Correlation as a Perceptual Image Quality Evaluator}, author={Zhu, Hanwei and Chen, Baoliang and Zhu, Lingyu and Wang, Shiqi and Weisi, Lin}, journal={arXiv preprint arXiv:2211.04927}, year={2024},

}

Reference url: https://github.com/h4nwei/DeepDC

Module Contents

pyiqa.archs.deepdc_arch.names

class pyiqa.archs.deepdc_arch.MultiVGGFeaturesExtractor(target_features=('conv1_2', 'conv2_2', 'conv3_4', 'conv4_4', 'conv5_4'), use_input_norm=True, requires_grad=False)

Bases: torch.nn.Module

Extract multiple VGG19 feature maps for DeepDC.

Parameters:

• target_features (tuple[str, ...]) – VGG layer names to return.

• use_input_norm (bool) – Whether to apply ImageNet normalization.

• requires_grad (bool) – Whether backbone parameters require gradients.

forward(x)

Extract requested VGG feature maps.

Parameters:

x (torch.Tensor) – Input tensor in range [0, 1] with shape (N, 3, H, W).

Returns:

Mapping from feature names to feature tensors.

Return type:

collections.OrderedDict[str, torch.Tensor]

class pyiqa.archs.deepdc_arch.DeepDC(features_to_compute=('conv1_2', 'conv2_2', 'conv3_4', 'conv4_4', 'conv5_4'))

Bases: torch.nn.Module

DeepDC full-reference IQA metric.

The score is derived from distance correlation between feature-space double-centered distance matrices.

Parameters:

features_to_compute (tuple[str, ...]) – VGG feature names used in the DeepDC aggregation.

forward(x, y)

Compute DeepDC quality score.

Parameters:

• x (torch.Tensor) – Distorted image tensor with shape (N, 3, H, W).

• y (torch.Tensor) – Reference image tensor with shape (N, 3, H, W).

Returns:

Quality score tensor with shape (N, 1).

Return type:

torch.Tensor

Distance_Correlation(matrix_A, matrix_B)

Compute distance correlation between matrix batches.

Parameters:

• matrix_A (torch.Tensor) – First matrix batch of shape (N, C, C).

• matrix_B (torch.Tensor) – Second matrix batch of shape (N, C, C).

Returns:

Correlation values with shape (N,).

Return type:

torch.Tensor

pyiqa.archs.deepdc_arch.prepare_image(image, resize=True)

Convert a PIL image to a 4D tensor for DeepDC demos.

Parameters:

• image (PIL.Image.Image) – Input image.

• resize (bool) – If True, resize shortest side to 256 when both sides are larger than 256.

Returns:

Tensor with shape (1, 3, H, W).

Return type:

torch.Tensor

pyiqa.archs.deepdc_arch.parser