Enum DenseNetVariant

Namespace

AiDotNet.Enums

Assembly

AiDotNet.dll

Specifies the DenseNet model variant.

public enum DenseNetVariant

Fields

Custom = 4

Custom DenseNet variant for testing with minimal layers.

Use this variant for unit tests to minimize construction time. Typically uses [2, 2, 2, 2] block configuration with small growth rate.

DenseNet121 = 0

DenseNet-121: [6, 12, 24, 16] layers per block (8M parameters).

The most commonly used variant, offering a good balance between accuracy and efficiency.

DenseNet169 = 1

DenseNet-169: [6, 12, 32, 32] layers per block (14M parameters).

A deeper variant with improved accuracy at the cost of more computation.

DenseNet201 = 2

DenseNet-201: [6, 12, 48, 32] layers per block (20M parameters).

A very deep variant for tasks requiring high accuracy.

DenseNet264 = 3

DenseNet-264: [6, 12, 64, 48] layers per block (33M parameters).

The deepest standard variant, offering maximum accuracy potential.

Remarks

DenseNet (Densely Connected Convolutional Networks) variants differ in their depth and computational requirements. Each variant has different numbers of layers per dense block.

For Beginners: The number in the variant name (e.g., DenseNet121) indicates the total number of layers in the network. Higher numbers mean deeper networks with potentially better accuracy but requiring more computation time and memory.