Losses Utils

class master_thesis.utils.LossesUtils

Bases: object

Utilities class containing loss-related methods.

static masked_l1(y_hat, y, mask, batch_mask=None, reduction='mean', weight=1)

static perceptual(y_hat, y, model_vgg, weight=1)

Computes the perceptual loss of the image y_hat with respect to the ground truth y.

Parameters

• y_hat – Tensor of size (B,C,H,W) containing the estimated image.

• y – Tensor of size (B,C,H,W) containing the ground-truth image.

• weight – Scaling factor applied to the loss.

Returns

Perceptual loss between y_hat and y.

static grad(y_hat, y, reduction, weight=1)

Computes the gradient loss of the image y_hat with respect to the ground truth y.

Parameters

• y_hat – Tensor of size (B,C,H,W) containing the estimated image.

• y – Tensor of size (B,C,H,W) containing the ground-truth image.

• reduction – Reduction mode applied to the loss.

• weight – Scaling factor applied to the loss.

Returns

Gradient loss between y_hat and y.