tensorplay.functional.quantized_batch_norm

tensorplay.functional.quantized_batch_norm(input, weight=None, bias=None, mean, var, eps, output_scale, output_zero_point) → Tensor[source]

Applies batch normalization on a 4D (NCHW) quantized tensor.

$$y = \frac{x - \mathrm{E}[x]}{\sqrt{\mathrm{Var}[x] + \epsilon}} * \gamma + \beta$$

Parameters:

• input (Tensor) – quantized tensor

• weight (Tensor) – float tensor that corresponds to the gamma, size C

• bias (Tensor) – float tensor that corresponds to the beta, size C

• mean (Tensor) – float mean value in batch normalization, size C

• var (Tensor) – float tensor for variance, size C

• eps (float) – a value added to the denominator for numerical stability.

• output_scale (float) – output quantized tensor scale

• output_zero_point (int) – output quantized tensor zero_point

Returns:

A quantized tensor with batch normalization applied.

Return type:

Tensor

Example:

>>> qx = tensorplay.quantize_per_tensor(tensorplay.rand(2, 2, 2, 2), 1.5, 3, tensorplay.quint8)
>>> tensorplay.quantized_batch_norm(qx, tensorplay.ones(2), tensorplay.zeros(2), tensorplay.rand(2), tensorplay.rand(2), 0.00001, 0.2, 2)
tensor([[[[-0.2000, -0.2000],
      [ 1.6000, -0.2000]],

     [[-0.4000, -0.4000],
      [-0.4000,  0.6000]]],


    [[[-0.2000, -0.2000],
      [-0.2000, -0.2000]],

     [[ 0.6000, -0.4000],
      [ 0.6000, -0.4000]]]], size=(2, 2, 2, 2), dtype=tensorplay.quint8,
   quantization_scheme=tensorplay.per_tensor_affine, scale=0.2, zero_point=2)