INNER CODE UNIT · Python
dmll
nv-tlabs/ATISS · scene_synthesis/losses/__init__.py:39
def dmll(pred, target, log_scale_min=-7.0, num_classes=256):
"""Discretized mixture of logistic distributions loss
Note that it is assumed that input is scaled to [-1, 1].
Code adapted
from https://github.com/idiap/linear-transformer-experiments/blob/0a540938ec95e1ec5b159ceabe0463d748ba626c/image-generation/utils.py#L31
Arguments
----------
pred (Tensor): Predicted output (B x L x T)
target (Tensor): Target (B x L x 1).
log_scale_min (float): Log scale minimum value
num_classes (int): Number of classes
Returns:
--------
Tensor: loss
"""