box_embeddings.modules.volume.tf_soft_volume

Module Contents

eps = 1e-23

tf_soft_volume(box_tensor: box_embeddings.parameterizations.tf_box_tensor.TFBoxTensor, volume_temperature: float = 1.0, log_scale: bool = True, scale: float = 1.0) → tensorflow.Tensor

Volume of boxes. Uses softplus instead of ReLU/clamp

Parameters

• box_tensor – input

• volume_temperature – 1/volume_temperature is the beta parameter for the softplus

• log_scale – Whether the output should be in log scale or not.

• scale – scale parameter. Should be left as 1.0 (default) in most cases.

Returns

Tensor of shape (…, ) when self has shape (…, 2, num_dims)

Raises

ValueError – if scale not in (0,1] or volume_temperature is 0.

class TFSoftVolume(log_scale: bool = True, volume_temperature: float = 1.0)

Bases: box_embeddings.modules.volume._tf_volume._TFVolume

Softplus based volume.