pyiqa.archs.wadiqam_arch

WaDIQaM model.

Reference:

Bosse, Sebastian, Dominique Maniry, Klaus-Robert Müller, Thomas Wiegand, and Wojciech Samek. “Deep neural networks for no-reference and full-reference image quality assessment.” IEEE Transactions on image processing 27, no. 1 (2017): 206-219.

Created by: https://github.com/lidq92/WaDIQaM Modified by: Chaofeng Chen (https://github.com/chaofengc) Refer to:

Official code from https://github.com/dmaniry/deepIQA

Module Contents

pyiqa.archs.wadiqam_arch.default_model_urls

pyiqa.archs.wadiqam_arch.make_layers(cfg: List[str | int]) → torch.nn.Sequential

class pyiqa.archs.wadiqam_arch.WaDIQaM(metric_type='FR', model_name='wadiqam_fr_kadid', pretrained=True, weighted_average=True, train_patch_num=32, pretrained_model_path=None, load_feature_weight_only=False, eps=1e-08)

Bases: torch.nn.Module

WaDIQaM model. :param metric_type: Choose metric mode. :type metric_type: String :param weighted_average: Average the weight. :type weighted_average: Boolean :param train_patch_num: Number of patch trained. Default: 32. :type train_patch_num: int :param pretrained_model_path: The pretrained model path. :type pretrained_model_path: String :param load_feature_weight_only: Only load featureweight. :type load_feature_weight_only: Boolean :param eps: Constant value. :type eps: float

load_pretrained_network(model_path, load_feature_weight_only=False)

get_patches(x, y=None)

extract_features(patches)

forward(x, y=None)

WaDIQaM model. :param x: An input tensor. Shape \((N, C, H, W)\). :param y: A reference tensor. Shape \((N, C, H, W)\).