tf.optimizers / schedules / schedules.Overview

TensorFlow 1 version

View source on GitHub

Dequantize the 'input' tensor into a float or bfloat16 Tensor.

tf.quantization.dequantize(
    input, min_range, max_range, mode='MIN_COMBINED', name=None,
    axis=None, narrow_range=False, dtype=tf.dtypes.float32
)

[min_range, max_range] are scalar floats that specify the range for the output. The 'mode' attribute controls exactly which calculations are used to convert the float values to their quantized equivalents.

In 'MIN_COMBINED' mode, each value of the tensor will undergo the following:

if T == qint8: in[i] += (range(T) + 1)/ 2.0
out[i] = min_range + (in[i]* (max_range - min_range) / range(T))

here range(T) = numeric_limits<T>::max() - numeric_limits<T>::min()

MIN_COMBINED Mode Example

If the input comes from a QuantizedRelu6, the output type is quint8 (range of 0-255) but the possible range of QuantizedRelu6 is 0-6. The min_range and max_range values are therefore 0.0 and 6.0. Dequantize on quint8 will take each value, cast to float, and multiply by 6 / 255. Note that if quantizedtype is qint8, the operation will additionally add each value by 128 prior to casting.

If the mode is 'MIN_FIRST', then this approach is used:

num_discrete_values = 1 << (# of bits in T)
range_adjust = num_discrete_values / (num_discrete_values - 1)
range = (range_max - range_min) * range_adjust
range_scale = range / num_discrete_values
const double offset_input = static_cast<double>(input) - lowest_quantized;
result = range_min + ((input - numeric_limits<T>::min()) * range_scale)

If the mode is SCALED, dequantization is performed by multiplying each input value by a scaling_factor. (Thus an input of 0 always maps to 0.0).

The scaling_factor is determined from min_range, max_range, and narrow_range in a way that is compatible with QuantizeAndDequantize{V2|V3} and QuantizeV2, using the following algorithm:

  const int min_expected_T = std::numeric_limits<T>::min() +
    (narrow_range ? 1 : 0);
  const int max_expected_T = std::numeric_limits<T>::max();
  const float max_expected_T = std::numeric_limits<float>::max();

  const float scale_factor =
    (std::numeric_limits<T>::min() == 0) ? (max_range / max_expected_T)
                                         : std::max(min_range / min_expected_T,
                                                    max_range / max_expected_T);