DenseNet

ConvNet

class lucid.models.DenseNet(config: DenseNetConfig)

The DenseNet class implements the Dense Convolutional Network family with dense feature reuse across each block. Model structure is defined through DenseNetConfig, which captures the dense block depths, growth rate, initial stem width, and classifier settings together with optional input-channel and transition-compression overrides.

        %%{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'>densenet_121</span>"]
style sg_m0 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    subgraph sg_m1["ConvBNReLU2d"]
    style sg_m1 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    m2["Conv2d<br/><span style='font-size:11px;color:#c53030;font-weight:400'>(1,3,224,224) → (1,64,112,112)</span>"];
    m3["BatchNorm2d"];
    m4["ReLU"];
    end
    m5["MaxPool2d<br/><span style='font-size:11px;color:#b7791f;font-weight:400'>(1,64,112,112) → (1,64,56,56)</span>"];
    subgraph sg_m6["blocks"]
    style sg_m6 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    m7(["_DenseBlock x 4<br/><span style='font-size:11px;font-weight:400'>(1,64,56,56) → (1,256,56,56)</span>"]);
    end
    subgraph sg_m8["transitions"]
    style sg_m8 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    m9(["_TransitionLayer x 3<br/><span style='font-size:11px;font-weight:400'>(1,256,56,56) → (1,128,28,28)</span>"]);
    end
    m10["BatchNorm2d"];
    m11["ReLU"];
    m12["AdaptiveAvgPool2d<br/><span style='font-size:11px;color:#b7791f;font-weight:400'>(1,1024,7,7) → (1,1024,1,1)</span>"];
    m13["Linear<br/><span style='font-size:11px;color:#2b6cb0;font-weight:400'>(1,1024) → (1,1000)</span>"];
end
input["Input<br/><span style='font-size:11px;color:#a67c00;font-weight:400'>(1,3,224,224)</span>"];
output["Output<br/><span style='font-size:11px;color:#a67c00;font-weight:400'>(1,1000)</span>"];
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 m10 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
style m11 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
style m12 fill:#fefcbf,stroke:#b7791f,stroke-width:1px;
style m13 fill:#ebf8ff,stroke:#2b6cb0,stroke-width:1px;
input --> m2;
m10 --> m11;
m11 --> m12;
m12 --> m13;
m13 --> output;
m2 --> m3;
m3 --> m4;
m4 --> m5;
m5 -.-> m7;
m7 --> m10;
m7 --> m9;
m9 -.-> m7;
    

Class Signature

class DenseNet(nn.Module):
    def __init__(self, config: DenseNetConfig)

Parameters

• config (DenseNetConfig): Configuration object describing the dense block depths, growth rate, stem width, classifier size, and optional input-channel, bottleneck, and compression settings.

Attributes

• config (DenseNetConfig): The configuration used to construct the model.

• growth_rate (int): Number of new channels contributed by each dense layer.

• num_init_features (int): Output width of the initial convolution stem.

• bottleneck (int): Expansion factor used by the 1x1 bottleneck convolution inside each dense layer.

• compression (float): Transition-layer compression ratio applied between dense blocks.

• conv0, pool0, blocks, transitions, bn_final, relu, avgpool, fc: DenseNet stem, dense blocks, transition layers, normalization head, and classifier.

Examples

>>> import lucid
>>> import lucid.models as models
>>> config = models.DenseNetConfig(
...     block_config=[6, 12, 24, 16],
...     growth_rate=32,
...     num_init_features=64,
...     num_classes=10,
...     in_channels=1,
... )
>>> model = models.DenseNet(config)
>>> output = model(lucid.zeros(1, 1, 224, 224))
>>> print(output.shape)
(1, 10)

Note

• Dense connectivity improves feature reuse and gradient flow across very deep stacks.

• Factory helpers such as densenet_121 and densenet_201 provide standard ImageNet presets.