rpn_prior¶
- probly.losses.torch.rpn_prior(shape: Size | tuple[int, ...], device: device) tuple[Tensor, Tensor, Tensor, Tensor][source]¶
Normal-Gamma prior for Regression Prior Networks from [MCPG20].
Constructs an uninformative Normal-Gamma prior used in Regression Prior Networks to regularize out-of-distribution predictions via KL divergence.
- Parameters:
shape – Shape of the prior parameter tensors (e.g., batch shape).
device – Torch device on which to allocate the tensors.
- Returns:
Tuple
(mu0, kappa0, alpha0, beta0)of Normal-Gamma prior parameters, each with the specified shape.