pyiqa.losses.iqa_losses¶

Module Contents¶

pyiqa.losses.iqa_losses.emd_loss(pred, target, r=2)[source]¶

Parameters:

pred (Tensor) – of shape (N, C). Predicted tensor.
target (Tensor) – of shape (N, C). Ground truth tensor.
r (float) – norm level, default l2 norm.

class pyiqa.losses.iqa_losses.EMDLoss(loss_weight=1.0, r=2, reduction='mean')[source]¶

Bases: torch.nn.Module

EMD (earth mover distance) loss.

forward(pred, target, weight=None, **kwargs)[source]¶

pyiqa.losses.iqa_losses.plcc_loss(pred, target)[source]¶

Parameters:

pred (Tensor) – of shape (N, 1). Predicted tensor.
target (Tensor) – of shape (N, 1). Ground truth tensor.

class pyiqa.losses.iqa_losses.PLCCLoss(loss_weight=1.0)[source]¶

Bases: torch.nn.Module

PLCC loss, induced from Pearson’s Linear Correlation Coefficient.

forward(pred, target)[source]¶

class pyiqa.losses.iqa_losses.RankLoss(detach=False, loss_weight=1.0)[source]¶

Bases: torch.nn.Module

Monotonicity regularization loss, will be zero when rankings of pred and target are the same.

Reference:

https://github.com/lidq92/LinearityIQA/blob/master/IQAloss.py

forward(pred, target)[source]¶

pyiqa.losses.iqa_losses.norm_loss_with_normalization(pred, target, p, q)[source]¶

Parameters:

pred (Tensor) – of shape (N, 1). Predicted tensor.
target (Tensor) – of shape (N, 1). Ground truth tensor.

class pyiqa.losses.iqa_losses.NiNLoss(loss_weight=1.0, p=1, q=2)[source]¶

Bases: torch.nn.Module

NiN (Norm in Norm) loss

Reference:

Dingquan Li, Tingting Jiang, and Ming Jiang. Norm-in-Norm Loss with Faster Convergence and Better Performance for Image Quality Assessment. ACMM2020.

https://arxiv.org/abs/2008.03889

https://github.com/lidq92/LinearityIQA

This loss can be simply described as: l1_norm(normalize(pred - pred_mean), normalize(target - target_mean))

forward(pred, target)[source]¶