torch.js has not been released yet.

torch.nn.InstanceNorm3d

class InstanceNorm3d extends _InstanceNorm

Instance Normalization for 3D inputs (volumetric): normalizes each volume independently.

Per-sample normalization for 3D/volumetric data. Normalizes each sample independently across depth, height, width. Essential for 3D generative models and volumetric processing.

\begin{aligned} \mu_{n,c} = \frac{1}{D \cdot H \cdot W} \sum_{d,h,w} x[n,c,d,h,w], \quad \sigma_{n,c}^2 = \frac{1}{D \cdot H \cdot W} \sum_{d,h,w} (x[n,c,d,h,w] - \mu_{n,c})^2 \\ x_{\text{norm}} = \frac{x - \mu_{n,c}}{\sqrt{\sigma_{n,c}^2 + \epsilon}}, \quad y = \gamma x_{\text{norm}} + \beta \end{aligned}

3D volumetric normalization: Normalizes across depth, height, width
Per-sample: Each 3D volume treated independently
Memory intensive: 3D operations use significant memory

Examples

// 3D generative model with instance normalization
const norm = new torch.nn.InstanceNorm3d(128);
const x = torch.randn([4, 128, 32, 32, 32]);  // [batch, channels, D, H, W]
const y = norm.forward(x);  // Each volume normalized independently

See Also

PyTorch torch.nn.InstanceNorm3d

InstanceNormOptions