inception_resnet_v2

lucid.models.inception_resnet_v2(num_classes: int = 1000, **kwargs) → InceptionResNet

Overview

The inception_resnet_v2 function implements the Inception-ResNet v2 architecture, which builds on Inception-ResNet v1 with deeper layers and improved efficiency. This model leverages the advantages of Inception modules and residual connections for enhanced performance on complex image classification tasks.

This function returns a preconfigured InceptionResNet model optimized for use in various applications, while allowing for flexibility in the number of output classes and other customizations. It builds an InceptionResNetConfig preset internally and forwards extra keyword arguments to that config.

Total Parameters: 35,847,512

Function Signature

@register_model
def inception_resnet_v2(
    num_classes: int = 1000,
    **kwargs,
) -> InceptionResNet

Parameters

• num_classes (int, optional): The number of output classes for the final classification layer. Default is 1000.

• kwargs (dict, optional): Additional keyword arguments forwarded to InceptionResNetConfig, such as in_channels or dropout_prob.

Returns

• InceptionResNet: An instance of the InceptionResNet model configured for the v2 architecture.

Example Usage

Below is an example of defining and using the inception_resnet_v2 function:

import lucid.models as models

# Create an Inception-ResNet v2 model with default parameters
model = models.inception_resnet_v2(num_classes=1000)

# Sample input tensor (e.g., batch of 299x299 RGB images)
input_tensor = lucid.Tensor([...])

# Forward pass
output = model(input_tensor)
print(output)

model = models.inception_resnet_v2(
    num_classes=10,
    in_channels=1,
    dropout_prob=0.25,
)

print(model.config)