LrnLayer.md

CLrnLayer Class

• CLrnLayer Class
  • Settings
    • Window size
    • Bias
    • Scale (alpha)
    • Exponent (beta)
  • Trainable parameters
    • Final values
  • Inputs
  • Outputs

The class implements a layer that performs local response normalization using the following formula:

LRN(x)[obj][ch] = x[obj][ch] * / ((bias + alpha * sqrSum[obj][ch] / windowSize) ^ beta)

where:

• obj is index of object [0; BlobSize / Channels)
• ch is index of channel [0; Channels)
• windowSize, bias, alpha, beta are settings
• sqrSum is calculated using the following formula:

sqrSum(x)[obj][ch] = sum(x[obj][i] * x[obj][i] for each i in [ch_min, ch_max])
ch_min = max(0, ch - floor((windowSize - 1)/2))
ch_max = min(C - 1, ch + ceil((windowSize - 1)/2))

Settings

Window size

void SetWindowSize( int value );

Sets size of the window used during the calculation of sqrSum.

Bias

void SetBias( float value );

Sets the bias value, which is added to the scaled sum of squares.

Scale (alpha)

void SetAlpha( float value );

Sets the scale value. The sum of squares is multiplied by this value.

Exponent (beta)

void SetBeta( float value );

Sets the exponent, used in the formula.

Trainable parameters

There are no trainable parameters for this layer.

Inputs

The single input accepts a blob of any size.

Outputs

The single output contains a blob of the same size with the results of local response normalization.