fn
batch_norm_eval_op
→TensorImplPtrint batch_norm_eval_op(const int & x, const int & mean, const int & var, const int & gamma, const int & beta, double eps)Public free-function entry point for inference-mode Batch Normalization.
Thin wrapper that delegates to BatchNormEvalBackward::forward.
This is the symbol the pybind11 binding layer forwards to from
lucid.nn.functional.batch_norm when training=False.
See Also
BatchNormEvalBackward : Autograd node implementing the forward + backward. batch_norm_op : Training-mode counterpart that uses batch stats.
Parameters
xTensorImplPtrInput tensor of shape
(B, C, ...).meanTensorImplPtrRunning per-channel mean of shape
(C,).varTensorImplPtrRunning per-channel variance of shape
(C,).gammaTensorImplPtrPer-channel scale of shape
(C,).betaTensorImplPtrPer-channel shift of shape
(C,).epsdoubleNumerical-stability constant.
Returns
TensorImplPtrNormalised output, same shape as x.