FCN

ConvNet Segmentation ConvNet

class lucid.models.FCN(config: FCNConfig)

FCN is a fully convolutional semantic segmentation model built around a ResNet feature extractor and lightweight classifier heads. The backbone produces dense feature maps, the main classifier predicts per-class logits, and an optional auxiliary classifier provides intermediate supervision from a shallower stage.

        %%{init: {"flowchart":{"curve":"monotoneX","nodeSpacing":50,"rankSpacing":50}} }%%
flowchart LR
linkStyle default stroke-width:2.0px
subgraph sg_m0["<span style='font-size:20px;font-weight:700'>fcn_resnet_50</span>"]
style sg_m0 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    subgraph sg_m1["_FCNResNetBackbone"]
    direction TB;
    style sg_m1 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    m2["Conv2d"];
    m3["BatchNorm2d"];
    m4["ReLU"];
    m5["MaxPool2d"];
    subgraph sg_m6["layer1 x 4"]
        direction TB;
    style sg_m6 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
        m6_in(["Input"]);
        m6_out(["Output"]);
style m6_in fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
style m6_out fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
        m7(["_Bottleneck x 3"]);
    end
    end
    subgraph sg_m8["classifier x 2"]
    direction TB;
    style sg_m8 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    m8_in(["Input"]);
    m8_out(["Output"]);
style m8_in fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
style m8_out fill:#e2e8f0,stroke:#64748b,stroke-width:1px;
    m9["Conv2d"];
    m10["BatchNorm2d"];
    m11["ReLU"];
    m12["Dropout"];
    m13["Conv2d"];
    end
end
input["Input"];
output["Output"];
style input fill:#fff3cd,stroke:#a67c00,stroke-width:1px;
style output fill:#fff3cd,stroke:#a67c00,stroke-width:1px;
style m2 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m3 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m4 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m5 fill:#fefcbf,stroke:#b7791f,stroke-width:1px;
style m9 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
style m10 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m11 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m12 fill:#edf2f7,stroke:#4a5568,stroke-width:1px;
style m13 fill:#ffe8e8,stroke:#c53030,stroke-width:1px;
input --> m2;
m10 --> m11;
m11 --> m12;
m12 --> m13;
m13 --> m8_in;
m13 --> m8_out;
m2 --> m3;
m3 --> m4;
m4 --> m5;
m5 -.-> m7;
m6_in -.-> m7;
m6_out -.-> m6_in;
m6_out -.-> m9;
m7 -.-> m6_in;
m7 --> m6_out;
m8_in -.-> m9;
m8_out --> output;
m9 --> m10;
    

Class Signature

class FCN(_FCNSegmentationModel, PreTrainedModelMixin):
    def __init__(self, config: FCNConfig) -> None

Parameters

• config (FCNConfig): Model configuration describing the backbone choice, classifier widths, and segmentation output space.

Methods

FCN.forward(x: Tensor, return_aux: bool = False) → Tensor | dict[str, Tensor]

Examples

Build from a Preset Builder

from lucid.models import fcn_resnet_50
import lucid

model = fcn_resnet_50(num_classes=21)
x = lucid.random.randn(1, 3, 224, 224)

logits = model(x)
print(logits.shape)

Return Auxiliary Output

from lucid.models import fcn_resnet_50
import lucid

model = fcn_resnet_50(num_classes=21, aux_loss=True)
x = lucid.random.randn(1, 3, 224, 224)

out = model(x, return_aux=True)
print(out["out"].shape)
print(out["aux"].shape)