Transformer

MLP

class alonet.transformers.mlp.MLP(input_dim, hidden_dim, output_dim, num_layers)

Bases: torch.nn.modules.module.Module

Very simple multi-layer perceptron (also called FFN)

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

Positional Encoding

Various positional encodings for the transformer.

class alonet.transformers.position_encoding.PositionEmbeddingSine(num_pos_feats=64, temperature=10000, normalize=False, scale=None, center=False)

Bases: torch.nn.modules.module.Module

This is a more standard version of the position embedding, very similar to the one used by the Attention is all you need paper, generalized to work on images.

forward(ftmap_mask)

TODO. We must made one global position encoding in the transformer class and one specific class in Detr that inhert from the main psoition embedding method.

Parameters

ftmap_mask: tuple

Tuple made of one torch.Tensor (The feature map) and one mask for the transformer

training: bool

alonet.transformers.position_encoding.build_position_encoding()

Transformer