l5kit.planning.vectorized.local_graph module¶
- class l5kit.planning.vectorized.local_graph.LocalMLP(dim_in: int, use_norm: bool = True)¶
Bases:
torch.nn.modules.module.Module- forward(x: torch.Tensor) torch.Tensor¶
forward of the module
- Parameters
x (torch.Tensor) – input tensor (…, dim_in)
- Returns
output tensor (…, dim_in)
- Return type
torch.Tensor
- training: bool¶
- class l5kit.planning.vectorized.local_graph.LocalSubGraph(num_layers: int, dim_in: int)¶
Bases:
torch.nn.modules.module.Module- forward(x: torch.Tensor, invalid_mask: torch.Tensor, pos_enc: torch.Tensor) torch.Tensor¶
Forward of the module: - Add positional encoding - Forward to layers - Aggregates using max (calculates a feature descriptor per element - reduces over points)
- Parameters
x (torch.Tensor) – input tensor (B,N,P,dim_in)
invalid_mask (torch.Tensor) – invalid mask for x (B,N,P)
pos_enc (torch.Tensor) – positional_encoding for x
- Returns
output tensor (B,N,P,dim_in)
- Return type
torch.Tensor
- training: bool¶
- class l5kit.planning.vectorized.local_graph.LocalSubGraphLayer(dim_in: int, dim_out: int)¶
Bases:
torch.nn.modules.module.Module- forward(x: torch.Tensor, invalid_mask: torch.Tensor) torch.Tensor¶
Forward of the model
- Parameters
x – input tensor
:tensor (B,N,P,dim_in) :param invalid_mask: invalid mask for x :tensor invalid_mask (B,N,P) :return: output tensor (B,N,P,dim_out) :rtype: torch.Tensor
- training: bool¶