pyiqa.archs.arniqa_arch

ARNIQA: Learning Distortion Manifold for Image Quality Assessment

@inproceedings{agnolucci2024arniqa,

title={ARNIQA: Learning Distortion Manifold for Image Quality Assessment}, author={Agnolucci, Lorenzo and Galteri, Leonardo and Bertini, Marco and Del Bimbo, Alberto}, booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision}, pages={189–198}, year={2024}

}

Reference:

• Arxiv link: https://www.arxiv.org/abs/2310.14918

• Official Github: https://github.com/miccunifi/ARNIQA

Module Contents

pyiqa.archs.arniqa_arch.DATASET_INFO

pyiqa.archs.arniqa_arch.default_model_urls

class pyiqa.archs.arniqa_arch.ARNIQA(regressor_dataset: str = 'koniq')

Bases: torch.nn.Module

ARNIQA model implementation.

This class implements the ARNIQA model for image quality assessment, which combines a ResNet50 encoder with a regressor network for predicting image quality scores.

Parameters:

regressor_dataset (str, optional) – The dataset to use for the regressor. Default is “koniq”.

regressor_dataset

The dataset to use for the regressor.

Type:

str

encoder

The ResNet50 encoder.

Type:

nn.Module

feat_dim

The feature dimension of the encoder.

Type:

int

regressor

The regressor network.

Type:

nn.Module

default_mean

The mean values for normalization.

Type:

torch.Tensor

default_std

The standard deviation values for normalization.

Type:

torch.Tensor

forward(x: torch.Tensor) → float

Forward pass of the ARNIQA model.

Parameters:

x (torch.Tensor) – The input tensor.

Returns:

The predicted quality score.

Return type:

float