Mixed Quantization

Per-Channel Quantization

Quantization technique using distinct parameters for each channel or group of channels in a convolutional layer. This method better preserves precision by adapting scale to specific characteristics of each filter.

← Tillbaka