ImpactLoss.h File Reference

#include <torch/torch.h>
#include <cmath>
#include <iostream>
#include "itkTimeProbe.h"

Include dependency graph for ImpactLoss.h:

Go to the source code of this file.

Classes
class	ImpactLoss::Cosine
class	ImpactLoss::Dice
class	ImpactLoss::DotProduct
class	ImpactLoss::L1
class	ImpactLoss::L1Cosine
class	ImpactLoss::L2
class	ImpactLoss::Loss
class	ImpactLoss::LossFactory
class	ImpactLoss::NCC
class	ImpactLoss::RegisterLoss< T >

Namespaces
namespace	ImpactLoss

Functions
RegisterLoss< Cosine >	ImpactLoss::CosineReg ("Cosine")
RegisterLoss< Dice >	ImpactLoss::Dice_reg ("Dice")
RegisterLoss< DotProduct >	ImpactLoss::DotProductReg ("DotProduct")
RegisterLoss< L1 >	ImpactLoss::L1_reg ("L1")
RegisterLoss< L1Cosine >	ImpactLoss::L1CosineReg ("L1Cosine")
RegisterLoss< L2 >	ImpactLoss::MSE_reg ("L2")
RegisterLoss< NCC >	ImpactLoss::NCC_reg ("NCC")

Detailed Description

Implementation of differentiable loss functions used in the IMPACT image registration metric. Each loss is implemented as a class inheriting from Loss, and can be registered dynamically via the LossFactory for use at runtime.

Supports both static and Jacobian-based backpropagation modes.

Author

V. Boussot, Univ. Rennes, INSERM, LTSI- UMR 1099, F-35000 Rennes, France

Note

This work was funded by the French National Research Agency as part of the VATSop project (ANR-20-CE19-0015).

If you use the Impact anywhere we would appreciate if you cite the following article:
V. Boussot et al., IMPACT: A Generic Semantic Loss for Multimodal Medical Image Registration, arXiv preprint arXiv:2503.24121 (2025). https://doi.org/10.48550/arXiv.2503.24121

Definition in file ImpactLoss.h.