Skip to content
Calcrivo

CNN Parameter Count Calculator

Calculate the number of trainable parameters in a convolutional layer.

Inputs

Total Parameters

73,856

Weight Parameters

73,728

Step by step

  1. Weights: K² × C_in × C_out

    3² × 64 × 128

    = 73,728

  2. Biases

    128

    = 128

  3. Total

    73,728 + 128

    = 73,856

How it works

A convolutional layer has K×K×C_in×C_out weight parameters (kernel size squared times input channels times output filters) plus C_out bias terms. Conv layers are far more parameter-efficient than dense layers due to weight sharing across spatial positions.

Formula

Conv2D Parameters

params = kernel_size^2 × in_channels × out_channels + out_channels

kernel_size
Height/width of the convolution kernel

Frequently Asked Questions

Why are conv layers more efficient than dense layers?

A conv layer reuses the same small kernel across all spatial positions (weight sharing), requiring far fewer parameters than connecting every input pixel to every output neuron.

Does kernel size affect parameter count?

Yes — parameters scale with K². A 5×5 kernel has 2.78× more parameters than a 3×3 kernel. This is why modern architectures prefer stacking multiple 3×3 layers over using larger kernels.

You might also need