| 106 | # Lightly adapted from |
| 107 | # https://github.com/facebookresearch/MaskFormer/blob/main/mask_former/modeling/transformer/transformer_predictor.py # noqa |
| 108 | class MLP(nn.Module): |
| 109 | def __init__( |
| 110 | self, |
| 111 | input_dim: int, |
| 112 | hidden_dim: int, |
| 113 | output_dim: int, |
| 114 | num_layers: int, |
| 115 | activation: nn.Module = nn.ReLU, |
| 116 | sigmoid_output: bool = False, |
| 117 | ) -> None: |
| 118 | super().__init__() |
| 119 | self.num_layers = num_layers |
| 120 | h = [hidden_dim] * (num_layers - 1) |
| 121 | self.layers = nn.ModuleList( |
| 122 | nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim]) |
| 123 | ) |
| 124 | self.sigmoid_output = sigmoid_output |
| 125 | self.act = activation() |
| 126 | |
| 127 | def forward(self, x): |
| 128 | for i, layer in enumerate(self.layers): |
| 129 | x = self.act(layer(x)) if i < self.num_layers - 1 else layer(x) |
| 130 | if self.sigmoid_output: |
| 131 | x = F.sigmoid(x) |
| 132 | return x |
| 133 | |
| 134 | |
| 135 | # From https://github.com/facebookresearch/detectron2/blob/main/detectron2/layers/batch_norm.py # noqa |
no outgoing calls
no test coverage detected