maxvit_xlarge

lucid.models.maxvit_xlarge(in_channels: int = 3, num_classes: int = 1000, **kwargs) → MaxViT

The maxvit_xlarge function returns the highest capacity variant of the MaxViT architecture. It significantly increases both depth and channel width, making it ideal for high-resolution datasets and demanding image classification tasks.

Total Parameters: 383,734,024

Function Signature

def maxvit_xlarge(
    in_channels: int = 3,
    num_classes: int = 1000,
    **kwargs
) -> MaxViT

Parameters

  • in_channels (int, optional): Number of input channels (e.g., 3 for RGB images). Default is 3.

  • num_classes (int, optional): Number of output classes for classification. Default is 1000.

  • kwargs (any, optional): Additional keyword arguments passed to the MaxViT constructor.

Model Configuration

This xlarge variant is configured as:

  • depths: (2, 6, 14, 2)

  • channels: (192, 384, 768, 1536)

  • embed_dim: 192

These values offer maximum feature capacity, especially suitable for large-scale datasets.

Example

import lucid
from lucid.models.transformer import maxvit_xlarge

model = maxvit_xlarge()
input_tensor = lucid.randn(1, 3, 224, 224)
output = model(input_tensor)
print(output.shape)  # (1, 1000)

Warning

Due to its large number of parameters, this model is recommended for use on powerful hardware (e.g., GPU or Apple Silicon with MLX backend).