Implementation:Kornia Kornia SSIM3D Loss

Knowledge Sources	Kornia
Domains	Vision, Loss_Functions
Last Updated	2026-02-09 15:00 GMT

Overview

SSIM3D Loss computes a loss based on the 3D Structural Similarity Index Measure (SSIM) for volumetric data.

Description

The SSIM3D Loss extends the Structural Similarity Index Measure to 3D volumetric data. It computes the Structural Dissimilarity as:

$\text{loss}(x,y)=1-\text{SSIM}(x,y)$

Note that unlike the 2D SSIM loss which divides by 2, the 3D variant directly uses the complement of the SSIM value.

The SSIM3D computation uses a 3D Gaussian kernel for smoothing the volume data before computing similarity. It delegates to `kornia.metrics.ssim3d` for the actual SSIM computation.

The implementation supports two padding modes:

• same: Pads the input to produce output of the same spatial size.
• valid: Uses only the valid convolution area.

Usage

Import this loss for 3D medical image analysis tasks, video processing, and volumetric data reconstruction where you need a perceptual quality metric that operates on 3D data (e.g., CT scans, MRI volumes, video sequences).

Code Reference

Source Location

• Repository: Kornia
• File: kornia/losses/ssim3d.py
• Lines: 1-129

Signature

def ssim3d_loss(
    img1: torch.Tensor,
    img2: torch.Tensor,
    window_size: int,
    max_val: float = 1.0,
    eps: float = 1e-12,
    reduction: str = "mean",
    padding: str = "same",
) -> torch.Tensor: ...

class SSIM3DLoss(nn.Module):
    def __init__(
        self,
        window_size: int,
        max_val: float = 1.0,
        eps: float = 1e-12,
        reduction: str = "mean",
        padding: str = "same",
    ) -> None: ...
    def forward(self, img1: torch.Tensor, img2: torch.Tensor) -> torch.Tensor: ...

Import

from kornia.losses import SSIM3DLoss
from kornia.losses import ssim3d_loss

I/O Contract

Inputs

Name	Type	Required	Description
window_size	int	Yes	Size of the 3D Gaussian kernel for smoothing
max_val	float	No	Dynamic range of the images (default: 1.0)
eps	float	No	Small value for numerical stability (default: 1e-12)
reduction	str	No	Reduction mode: 'none', 'mean' (default), or 'sum'
padding	str	No	Padding mode: 'same' (default) or 'valid'
img1	torch.Tensor	Yes	First input volume with shape (B, C, D, H, W)
img2	torch.Tensor	Yes	Second input volume with shape (B, C, D, H, W)

Outputs

Name	Type	Description
loss	torch.Tensor	3D SSIM loss value (1 - SSIM); scalar for 'mean'/'sum', per-voxel for 'none'

Usage Examples

import torch
from kornia.losses import SSIM3DLoss

# Create sample 3D volume tensors
input1 = torch.rand(1, 4, 5, 5, 5)
input2 = torch.rand(1, 4, 5, 5, 5)

# Using the module API
criterion = SSIM3DLoss(window_size=5)
loss = criterion(input1, input2)

# With custom parameters
criterion_custom = SSIM3DLoss(
    window_size=7,
    max_val=1.0,
    reduction="mean",
    padding="valid",
)
loss = criterion_custom(input1, input2)

Related Pages

• Principle:Kornia_Kornia_Perceptual_Quality_Loss