FC

Fully connected layer head for classification. You can adjust the number of layers and channel sizes. Channel of last layer is always same with the number of classes.

Compatibility matrix

Supporting backbones	Supporting necks	torch.fx	NetsPresso
ResNet MobileNetV3 MixNet CSPDarkNet ViT MobileViT MixTransformer EfficientFormer		Supported	Supported

Field list

Field	Description
name	(str) Name must be "fc" to use FC head.
params.num_layers	(int) The number of fully connected layers. Channel of last layer is same with the number of classes.
params.intermediate_channels	(int) Dimension of intermediate fully connected layers. This can be ignored if num_layer is 1.
params.act_type	(str) Activation function for intermediate fully connected layers. This can be ignored if num_layer is 1.
params.dropout_prob	(float) Dropout probability before the last classifier layer.

Model configuration example

2-layer fully connected layer classifier

model:
  architecture:
    head:
      name: fc
      params:
        num_layers: 2
        intermediate_channels: 1024
        act_type: hard_swish
        dropout_prob: 0.2

Related links