nntoolbox.vision.losses.robust module

More robust loss functions (UNTESTED)

class nntoolbox.vision.losses.robust.CharbonnierLoss(eps: float = 0.001)

Bases: nntoolbox.vision.losses.robust.GeneralizedCharbonnierLoss

Charbonnier Loss Function:

l(input, target) = sqrt((input - target)^2 + eps^2)

References:

Wei-Sheng Lai et al. “Fast and Accurate Image Super-Resolution with Deep Laplacian Pyramid Networks.” https://arxiv.org/pdf/1710.01992.pdf

training: bool

class nntoolbox.vision.losses.robust.CharbonnierLossV2(eps: float = 0.001)

Bases: torch.nn.modules.module.Module

Charbonnier Loss Function:

l(input, target) = sqrt((input - target)^2 + eps^2)

References:

Wei-Sheng Lai et al. “Fast and Accurate Image Super-Resolution with Deep Laplacian Pyramid Networks.” https://arxiv.org/pdf/1710.01992.pdf

forward(input: torch.Tensor, target: torch.Tensor) → torch.Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class nntoolbox.vision.losses.robust.GeneralizedCharbonnierLoss(alpha: float = 1.0, eps: float = 1e-06)

Bases: torch.nn.modules.module.Module

Generalized Charbonnier Loss Function:

l(input, target) = (input - target)^2 + eps^2) ^ (alpha / 2)

References:

Deqing Sun et al. “Secrets of Optical Flow Estimation and Their Principles.” http://cs.brown.edu/~dqsun/pubs/cvpr_2010_flow.pdf

forward(input: torch.Tensor, target: torch.Tensor) → torch.Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool