ObjectNormalizationLayer.md

CObjectNormalizationLayer Class

• CObjectNormalizationLayer Class
  • Settings
    • Epsilon
  • Trainable parameters
    • Scale
    • Bias
  • Inputs
  • Outputs

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

objectNorm(x)[i][j] = ((x[i][j] - mean[i]) / sqrt(var[i] + epsilon)) * scale[j] + bias[j]

where:

• scale and bias are trainable parameters
• mean and var are the mean and variance of each object in a batch

Settings

Epsilon

void SetEpsilon( float newEpsilon );

Sets epsilon which is added to the variance in order to avoid division by zero.

Trainable parameters

Scale

CPtr<CDnnBlob> GetScale() const;

Gets the scale vector. It is a blob of any shape and of total size equal Height * Width * Depth * Channels of the input.

Bias

CPtr<CDnnBlob> GetBias() const;

Gets the bias vector. It is a blob of any shape and of total size equal Height * Width * Depth * Channels of the input.

Inputs

The single input accepts a blob containing BatchLength * BatchWidth * ListSize objects of size Height * Width * Depth * Channels.

Outputs

The single output contains a blob with the results, of the same size as the input blob.