InceptionNeXt

ConvNet

class lucid.models.InceptionNeXt(config: InceptionNeXtConfig)

The InceptionNeXt class implements a modernized version of the Inception architecture, leveraging depthwise convolutions, multi-layer perceptrons (MLPs), and advanced token mixing techniques for enhanced performance and efficiency. Model structure is defined through InceptionNeXtConfig.

        %%{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'>inception_next_base</span>"]
  style sg_m0 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
    subgraph sg_m1["stem"]
      direction TB;
    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) → (1,128,56,56)</span>"];
      m3["BatchNorm2d"];
    end
    subgraph sg_m4["stages"]
      direction TB;
    style sg_m4 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
      subgraph sg_m5["_IncepNeXtStage"]
        direction TB;
      style sg_m5 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
        m6["Identity"];
        m7["Sequential"];
      end
      subgraph sg_m8["_IncepNeXtStage x 3"]
        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(["Sequential x 2<br/><span style='font-size:11px;font-weight:400'>(1,128,56,56) → (1,256,28,28)</span>"]);
      end
    end
    subgraph sg_m10["_MLPHead"]
      direction TB;
    style sg_m10 fill:#000000,fill-opacity:0.05,stroke:#000000,stroke-opacity:0.75,stroke-width:1px
      m11["Linear<br/><span style='font-size:11px;color:#2b6cb0;font-weight:400'>(1,1024) → (1,3072)</span>"];
      m12["GELU"];
      m13["LayerNorm"];
      m14["AdaptiveAvgPool2d<br/><span style='font-size:11px;color:#b7791f;font-weight:400'>(1,1024,7,7) → (1,1024,1,1)</span>"];
      m15["Dropout"];
      m16["Linear<br/><span style='font-size:11px;color:#2b6cb0;font-weight:400'>(1,3072) → (1,1000)</span>"];
    end
  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 m6 fill:#ebf8ff,stroke:#2b6cb0,stroke-width:1px;
  style m11 fill:#ebf8ff,stroke:#2b6cb0,stroke-width:1px;
  style m12 fill:#faf5ff,stroke:#6b46c1,stroke-width:1px;
  style m13 fill:#e6fffa,stroke:#2c7a7b,stroke-width:1px;
  style m14 fill:#fefcbf,stroke:#b7791f,stroke-width:1px;
  style m15 fill:#edf2f7,stroke:#4a5568,stroke-width:1px;
  style m16 fill:#ebf8ff,stroke:#2b6cb0,stroke-width:1px;
  input --> m2;
  m11 --> m12;
  m12 --> m13;
  m13 --> m15;
  m14 --> m11;
  m15 --> m16;
  m16 --> output;
  m2 --> m3;
  m3 --> m6;
  m6 --> m7;
  m7 -.-> m9;
  m8_in -.-> m9;
  m8_out --> m14;
  m8_out -.-> m8_in;
  m9 -.-> m8_in;
  m9 --> m8_out;
    

Class Signature

class InceptionNeXt(nn.Module):
    def __init__(self, config: InceptionNeXtConfig) -> None

Parameters

• config (InceptionNeXtConfig): Configuration object describing stage depths, stage widths, token mixers, MLP ratios, classifier head settings, and regularization values.

Examples

from lucid.models import InceptionNeXt, InceptionNeXtConfig

config_tiny = InceptionNeXtConfig(
    num_classes=1000,
    depths=(3, 3, 9, 3),
    dims=(96, 192, 384, 768),
)
model_tiny = InceptionNeXt(config_tiny)

input_tensor = lucid.random.randn(1, 3, 224, 224)
output = model_tiny(input_tensor)
print(output.shape)