pytorch3d.implicitron.models.global_encoder.autodecoder

autodecoder

class pytorch3d.implicitron.models.global_encoder.autodecoder.Autodecoder(*args, **kwargs)

Bases: Configurable, Module

Autodecoder which maps a list of integer or string keys to optimizable embeddings.

Settings:

encoding_dim: Embedding dimension for the decoder. n_instances: The maximum number of instances stored by the autodecoder. init_scale: Scale factor for the initial autodecoder weights. ignore_input: If True, optimizes a single code for any input.

encoding_dim: int = 0

n_instances: int = 1

init_scale: float = 1.0

ignore_input: bool = False

calculate_squared_encoding_norm() → Tensor | None

get_encoding_dim() → int

forward(x: LongTensor | List[str]) → Tensor | None

Parameters:

• x – A batch of N identifiers. Either a long tensor of size

• ` (N,)` keys in [0, n_instances)

• codes (are hashed to)

Returns:

codes –

A tensor of shape (N, self.encoding_dim) containing the

key-specific autodecoder codes.